liz/rawTherapee
liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
rawTherapee/rtengine/FTblockDN.cc
Ingo Weyrich 033d9fe02a Merge branch 'dev' into newlocallab
2019-11-04 23:01:33 +01:00

3703 lines
153 KiB
C++
Raw Blame History

////////////////////////////////////////////////////////////////
//
// CFA denoise by wavelet transform, FT filtering
//
// copyright (c) 2008-2012 Emil Martinec <ejmartin@uchicago.edu>
//
//
// code dated: March 9, 2012
//
// FTblockDN.cc is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <https://www.gnu.org/licenses/>.
//
////////////////////////////////////////////////////////////////
#include <cmath>
#include <fftw3.h>
#include "array2D.h"
#include "boxblur.h"
#include "cplx_wavelet_dec.h"
#include "color.h"
#include "curves.h"
#include "iccmatrices.h"
#include "iccstore.h"
#include "imagefloat.h"
#include "improcfun.h"
#include "labimage.h"
#include "LUT.h"
#include "median.h"
#include "mytime.h"
#include "opthelper.h"
#include "procparams.h"
#include "rt_math.h"
#include "sleef.h"
#include "../rtgui/threadutils.h"
#include "../rtgui/options.h"
#ifdef _OPENMP
#include <omp.h>
#endif
//#define BENCHMARK
#include "StopWatch.h"
#define TS 64 // Tile size
#define offset 25 // shift between tiles
#define fTS ((TS/2+1)) // second dimension of Fourier tiles
#define blkrad 1 // radius of block averaging
#define epsilon 0.001f/(TS*TS) //tolerance
namespace rtengine
{
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
/*
Structure of the algorithm:
1. Compute an initial denoise of the image via undecimated wavelet transform
and universal thresholding modulated by user input.
2. Decompose the residual image into TSxTS size tiles, shifting by 'offset' each step
(so roughly each pixel is in (TS/offset)^2 tiles); Discrete Cosine transform the tiles.
3. Filter the DCT data to pick out patterns missed by the wavelet denoise
4. Inverse DCT the denoised tile data and combine the tiles into a denoised output image.
*/
extern MyMutex *fftwMutex;
namespace
{
template <bool useUpperBound>
void do_median_denoise(float **src, float **dst, float upperBound, int width, int height, ImProcFunctions::Median medianType, int iterations, int numThreads, float **buffer)
{
iterations = max(1, iterations);
typedef ImProcFunctions::Median Median;
int border = 1;
switch (medianType) {
case Median::TYPE_3X3_SOFT:
case Median::TYPE_3X3_STRONG: {
border = 1;
break;
}
case Median::TYPE_5X5_SOFT: {
border = 2;
break;
}
case Median::TYPE_5X5_STRONG: {
border = 2;
break;
}
case Median::TYPE_7X7: {
border = 3;
break;
}
case Median::TYPE_9X9: {
border = 4;
break;
}
}
float **allocBuffer = nullptr;
float **medBuffer[2];
medBuffer[0] = src;
// we need a buffer if src == dst or if (src != dst && iterations > 1)
if (src == dst || iterations > 1) {
if (buffer == nullptr) { // we didn't get a buffer => create one
allocBuffer = new float*[height];
for (int i = 0; i < height; ++i) {
allocBuffer[i] = new float[width];
}
medBuffer[1] = allocBuffer;
} else { // we got a buffer => use it
medBuffer[1] = buffer;
}
} else { // we can write directly into destination
medBuffer[1] = dst;
}
float ** medianIn, ** medianOut = nullptr;
int BufferIndex = 0;
for (int iteration = 1; iteration <= iterations; ++iteration) {
medianIn = medBuffer[BufferIndex];
medianOut = medBuffer[BufferIndex ^ 1];
if (iteration == 1) { // upper border
for (int i = 0; i < border; ++i) {
for (int j = 0; j < width; ++j) {
medianOut[i][j] = medianIn[i][j];
}
}
}
#ifdef _OPENMP
#pragma omp parallel for num_threads(numThreads) if (numThreads>1) schedule(dynamic,16)
#endif
for (int i = border; i < height - border; ++i) {
int j = 0;
for (; j < border; ++j) {
medianOut[i][j] = medianIn[i][j];
}
switch (medianType) {
case Median::TYPE_3X3_SOFT: {
for (; j < width - border; ++j) {
if (!useUpperBound || medianIn[i][j] <= upperBound) {
medianOut[i][j] = median(
medianIn[i - 1][j],
medianIn[i][j - 1],
medianIn[i][j],
medianIn[i][j + 1],
medianIn[i + 1][j]
);
} else {
medianOut[i][j] = medianIn[i][j];
}
}
break;
}
case Median::TYPE_3X3_STRONG: {
for (; j < width - border; ++j) {
if (!useUpperBound || medianIn[i][j] <= upperBound) {
medianOut[i][j] = median(
medianIn[i - 1][j - 1],
medianIn[i - 1][j],
medianIn[i - 1][j + 1],
medianIn[i][j - 1],
medianIn[i][j],
medianIn[i][j + 1],
medianIn[i + 1][j - 1],
medianIn[i + 1][j],
medianIn[i + 1][j + 1]
);
} else {
medianOut[i][j] = medianIn[i][j];
}
}
break;
}
case Median::TYPE_5X5_SOFT: {
for (; j < width - border; ++j) {
if (!useUpperBound || medianIn[i][j] <= upperBound) {
medianOut[i][j] = median(
medianIn[i - 2][j],
medianIn[i - 1][j - 1],
medianIn[i - 1][j],
medianIn[i - 1][j + 1],
medianIn[i][j - 2],
medianIn[i][j - 1],
medianIn[i][j],
medianIn[i][j + 1],
medianIn[i][j + 2],
medianIn[i + 1][j - 1],
medianIn[i + 1][j],
medianIn[i + 1][j + 1],
medianIn[i + 2][j]
);
} else {
medianOut[i][j] = medianIn[i][j];
}
}
break;
}
case Median::TYPE_5X5_STRONG: {
#ifdef __SSE2__
for (; !useUpperBound && j < width - border - 3; j += 4) {
STVFU(
medianOut[i][j],
median(
LVFU(medianIn[i - 2][j - 2]),
LVFU(medianIn[i - 2][j - 1]),
LVFU(medianIn[i - 2][j]),
LVFU(medianIn[i - 2][j + 1]),
LVFU(medianIn[i - 2][j + 2]),
LVFU(medianIn[i - 1][j - 2]),
LVFU(medianIn[i - 1][j - 1]),
LVFU(medianIn[i - 1][j]),
LVFU(medianIn[i - 1][j + 1]),
LVFU(medianIn[i - 1][j + 2]),
LVFU(medianIn[i][j - 2]),
LVFU(medianIn[i][j - 1]),
LVFU(medianIn[i][j]),
LVFU(medianIn[i][j + 1]),
LVFU(medianIn[i][j + 2]),
LVFU(medianIn[i + 1][j - 2]),
LVFU(medianIn[i + 1][j - 1]),
LVFU(medianIn[i + 1][j]),
LVFU(medianIn[i + 1][j + 1]),
LVFU(medianIn[i + 1][j + 2]),
LVFU(medianIn[i + 2][j - 2]),
LVFU(medianIn[i + 2][j - 1]),
LVFU(medianIn[i + 2][j]),
LVFU(medianIn[i + 2][j + 1]),
LVFU(medianIn[i + 2][j + 2])
)
);
}
#endif
for (; j < width - border; ++j) {
if (!useUpperBound || medianIn[i][j] <= upperBound) {
medianOut[i][j] = median(
medianIn[i - 2][j - 2],
medianIn[i - 2][j - 1],
medianIn[i - 2][j],
medianIn[i - 2][j + 1],
medianIn[i - 2][j + 2],
medianIn[i - 1][j - 2],
medianIn[i - 1][j - 1],
medianIn[i - 1][j],
medianIn[i - 1][j + 1],
medianIn[i - 1][j + 2],
medianIn[i][j - 2],
medianIn[i][j - 1],
medianIn[i][j],
medianIn[i][j + 1],
medianIn[i][j + 2],
medianIn[i + 1][j - 2],
medianIn[i + 1][j - 1],
medianIn[i + 1][j],
medianIn[i + 1][j + 1],
medianIn[i + 1][j + 2],
medianIn[i + 2][j - 2],
medianIn[i + 2][j - 1],
medianIn[i + 2][j],
medianIn[i + 2][j + 1],
medianIn[i + 2][j + 2]
);
} else {
medianOut[i][j] = medianIn[i][j];
}
}
break;
}
case Median::TYPE_7X7: {
#ifdef __SSE2__
std::array<vfloat, 49> vpp ALIGNED16;
for (; !useUpperBound && j < width - border - 3; j += 4) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) {
vpp[kk] = LVFU(medianIn[i + ii][j + jj]);
}
}
STVFU(medianOut[i][j], median(vpp));
}
#endif
std::array<float, 49> pp;
for (; j < width - border; ++j) {
if (!useUpperBound || medianIn[i][j] <= upperBound) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) {
pp[kk] = medianIn[i + ii][j + jj];
}
}
medianOut[i][j] = median(pp);
} else {
medianOut[i][j] = medianIn[i][j];
}
}
break;
}
case Median::TYPE_9X9: {
#ifdef __SSE2__
std::array<vfloat, 81> vpp ALIGNED16;
for (; !useUpperBound && j < width - border - 3; j += 4) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) {
vpp[kk] = LVFU(medianIn[i + ii][j + jj]);
}
}
STVFU(medianOut[i][j], median(vpp));
}
#endif
std::array<float, 81> pp;
for (; j < width - border; ++j) {
if (!useUpperBound || medianIn[i][j] <= upperBound) {
for (int kk = 0, ii = -border; ii <= border; ++ii) {
for (int jj = -border; jj <= border; ++jj, ++kk) {
pp[kk] = medianIn[i + ii][j + jj];
}
}
medianOut[i][j] = median(pp);
} else {
medianOut[i][j] = medianIn[i][j];
}
}
for (; j < width; ++j) {
medianOut[i][j] = medianIn[i][j];
}
break;
}
}
for (; j < width; ++j) {
medianOut[i][j] = medianIn[i][j];
}
}
if (iteration == 1) { // lower border
for (int i = height - border; i < height; ++i) {
for (int j = 0; j < width; ++j) {
medianOut[i][j] = medianIn[i][j];
}
}
}
BufferIndex ^= 1; // swap buffers
}
if (medianOut != dst) {
#ifdef _OPENMP
#pragma omp parallel for num_threads(numThreads) if (numThreads>1)
#endif
for (int i = 0; i < height; ++i) {
for (int j = 0; j < width; ++j) {
dst[i][j] = medianOut[i][j];
}
}
}
if (allocBuffer != nullptr) { // we allocated memory, so let's free it now
for (int i = 0; i < height; ++i) {
delete[] allocBuffer[i];
}
delete[] allocBuffer;
}
}
} // namespace
void ImProcFunctions::Median_Denoise(float **src, float **dst, const int width, const int height, const Median medianType, const int iterations, const int numThreads, float **buffer)
{
do_median_denoise<false>(src, dst, 0.f, width, height, medianType, iterations, numThreads, buffer);
}
void ImProcFunctions::Median_Denoise(float **src, float **dst, float upperBound, const int width, const int height, const Median medianType, const int iterations, const int numThreads, float **buffer)
{
do_median_denoise<true>(src, dst, upperBound, width, height, medianType, iterations, numThreads, buffer);
}
void ImProcFunctions::Tile_calc(int tilesize, int overlap, int kall, int imwidth, int imheight, int &numtiles_W, int &numtiles_H, int &tilewidth, int &tileheight, int &tileWskip, int &tileHskip)
{
if (kall == 2) {
if (imwidth < tilesize) {
numtiles_W = 1;
tileWskip = imwidth;
tilewidth = imwidth;
} else {
numtiles_W = ceil((static_cast<float>(imwidth)) / (tilesize - overlap));
tilewidth = ceil((static_cast<float>(imwidth)) / (numtiles_W)) + overlap;
tilewidth += (tilewidth & 1);
tileWskip = tilewidth - overlap;
}
if (imheight < tilesize) {
numtiles_H = 1;
tileHskip = imheight;
tileheight = imheight;
} else {
numtiles_H = ceil((static_cast<float>(imheight)) / (tilesize - overlap));
tileheight = ceil((static_cast<float>(imheight)) / (numtiles_H)) + overlap;
tileheight += (tileheight & 1);
tileHskip = tileheight - overlap;
}
}
if (kall == 0) {
numtiles_W = 1;
tileWskip = imwidth;
tilewidth = imwidth;
numtiles_H = 1;
tileHskip = imheight;
tileheight = imheight;
}
// printf("Nw=%d NH=%d tileW=%d tileH=%d\n",numtiles_W,numtiles_H,tileWskip,tileHskip);
}
int denoiseNestedLevels = 1;
enum nrquality {QUALITY_STANDARD, QUALITY_HIGH};
void ImProcFunctions::RGB_denoise(int kall, Imagefloat * src, Imagefloat * dst, Imagefloat * calclum, float * ch_M, float *max_r, float *max_b, bool isRAW, const procparams::DirPyrDenoiseParams & dnparams, const double expcomp, const NoiseCurve & noiseLCurve, const NoiseCurve & noiseCCurve, float &nresi, float &highresi)
{
BENCHFUN
//#ifdef _DEBUG
MyTime t1e, t2e;
t1e.set();
//#endif
if (dnparams.luma == 0 && dnparams.chroma == 0 && !dnparams.median && !noiseLCurve && !noiseCCurve) {
//nothing to do; copy src to dst or do nothing in case src == dst
if (src != dst) {
src->copyData(dst);
}
if (calclum) {
delete calclum;
calclum = nullptr;
}
return;
}
MyMutex::MyLock lock(*fftwMutex);
const nrquality nrQuality = (dnparams.smethod == "shal") ? QUALITY_STANDARD : QUALITY_HIGH;//shrink method
const float qhighFactor = (nrQuality == QUALITY_HIGH) ? 1.f / static_cast<float>(settings->nrhigh) : 1.0f;
const bool useNoiseCCurve = (noiseCCurve && noiseCCurve.getSum() > 5.f);
const bool useNoiseLCurve = (noiseLCurve && noiseLCurve.getSum() >= 7.f);
const bool autoch = (settings->leveldnautsimpl == 1 && (dnparams.Cmethod == "AUT" || dnparams.Cmethod == "PRE")) || (settings->leveldnautsimpl == 0 && (dnparams.C2method == "AUTO" || dnparams.C2method == "PREV"));
float** lumcalc = nullptr;
float* lumcalcBuffer = nullptr;
float** ccalc = nullptr;
float* ccalcBuffer = nullptr;
bool ponder = false;
float ponderCC = 1.f;
if (settings->leveldnautsimpl == 1 && params->dirpyrDenoise.Cmethod == "PON") {
ponder = true;
ponderCC = 0.5f;
}
if (settings->leveldnautsimpl == 1 && params->dirpyrDenoise.Cmethod == "PRE") {
ponderCC = 0.5f;
}
if (settings->leveldnautsimpl == 0 && params->dirpyrDenoise.Cmethod == "PREV") {
ponderCC = 0.5f;
}
int metchoice = 0;
if (dnparams.methodmed == "Lonly") {
metchoice = 1;
} else if (dnparams.methodmed == "Lab") {
metchoice = 2;
} else if (dnparams.methodmed == "ab") {
metchoice = 3;
} else if (dnparams.methodmed == "Lpab") {
metchoice = 4;
}
const bool denoiseMethodRgb = (dnparams.dmethod == "RGB");
// init luma noisevarL
const float noiseluma = static_cast<float>(dnparams.luma);
const float noisevarL = (useNoiseLCurve && (denoiseMethodRgb || !isRAW)) ? static_cast<float>(SQR(((noiseluma + 1.0) / 125.0) * (10. + (noiseluma + 1.0) / 25.0))) : static_cast<float>(SQR((noiseluma / 125.0) * (1.0 + noiseluma / 25.0)));
const bool denoiseLuminance = (noisevarL > 0.00001f);
//printf("NL=%f \n",noisevarL);
if (useNoiseLCurve || useNoiseCCurve) {
int hei = calclum->getHeight();
int wid = calclum->getWidth();
TMatrix wprofi = ICCStore::getInstance()->workingSpaceMatrix(params->icm.workingProfile);
const float wpi[3][3] = {
{static_cast<float>(wprofi[0][0]), static_cast<float>(wprofi[0][1]), static_cast<float>(wprofi[0][2])},
{static_cast<float>(wprofi[1][0]), static_cast<float>(wprofi[1][1]), static_cast<float>(wprofi[1][2])},
{static_cast<float>(wprofi[2][0]), static_cast<float>(wprofi[2][1]), static_cast<float>(wprofi[2][2])}
};
lumcalcBuffer = new float[hei * wid];
lumcalc = new float*[(hei)];
for (int i = 0; i < hei; ++i) {
lumcalc[i] = lumcalcBuffer + (i * wid);
}
ccalcBuffer = new float[hei * wid];
ccalc = new float*[(hei)];
for (int i = 0; i < hei; ++i) {
ccalc[i] = ccalcBuffer + (i * wid);
}
float cn100Precalc = 0.f;
if (useNoiseCCurve) {
cn100Precalc = SQR(1.f + ponderCC * (4.f * noiseCCurve[100.f / 60.f]));
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for (int ii = 0; ii < hei; ++ii) {
for (int jj = 0; jj < wid; ++jj) {
float LLum, AAum, BBum;
float RL = calclum->r(ii, jj);
float GL = calclum->g(ii, jj);
float BL = calclum->b(ii, jj);
// determine luminance and chrominance for noisecurves
float XL, YL, ZL;
Color::rgbxyz(RL, GL, BL, XL, YL, ZL, wpi);
Color::XYZ2Lab(XL, YL, ZL, LLum, AAum, BBum);
if (useNoiseLCurve) {
float epsi = 0.01f;
if (LLum < 2.f) {
LLum = 2.f; //avoid divided by zero
}
if (LLum > 32768.f) {
LLum = 32768.f; // not strictly necessary
}
float kinterm = epsi + noiseLCurve[xdivf(LLum, 15) * 500.f];
kinterm *= 100.f;
kinterm += noiseluma;
lumcalc[ii][jj] = SQR((kinterm / 125.f) * (1.f + kinterm / 25.f));
}
if (useNoiseCCurve) {
float cN = sqrtf(SQR(AAum) + SQR(BBum));
if (cN > 100) {
ccalc[ii][jj] = SQR(1.f + ponderCC * (4.f * noiseCCurve[cN / 60.f]));
} else {
ccalc[ii][jj] = cn100Precalc;
}
}
}
}
delete calclum;
calclum = nullptr;
}
const short int imheight = src->getHeight(), imwidth = src->getWidth();
if (dnparams.luma != 0 || dnparams.chroma != 0 || dnparams.methodmed == "Lab" || dnparams.methodmed == "Lonly") {
// gamma transform for input data
float gam = dnparams.gamma;
float gamthresh = 0.001f;
if (!isRAW) {//reduce gamma under 1 for Lab mode ==> TIF and JPG
if (gam < 1.9f) {
gam = 1.f - (1.9f - gam) / 3.f; //minimum gamma 0.7
} else if (gam >= 1.9f && gam <= 3.f) {
gam = (1.4f / 1.1f) * gam - 1.41818f;
}
}
LUTf gamcurve(65536, LUT_CLIP_BELOW);
float gamslope = exp(log(static_cast<double>(gamthresh)) / gam) / gamthresh;
if (denoiseMethodRgb) {
Color::gammaf2lut(gamcurve, gam, gamthresh, gamslope, 65535.f, 32768.f);
} else {
Color::gammanf2lut(gamcurve, gam, 65535.f, 32768.f);
}
// inverse gamma transform for output data
float igam = 1.f / gam;
float igamthresh = gamthresh * gamslope;
float igamslope = 1.f / gamslope;
LUTf igamcurve(65536, LUT_CLIP_BELOW);
if (denoiseMethodRgb) {
Color::gammaf2lut(igamcurve, igam, igamthresh, igamslope, 32768.f, 65535.f);
} else {
Color::gammanf2lut(igamcurve, igam, 32768.f, 65535.f);
}
const float gain = pow(2.0f, float(expcomp));
float params_Ldetail = min(float(dnparams.Ldetail), 99.9f); // max out to avoid div by zero when using noisevar_Ldetail as divisor
float noisevar_Ldetail = SQR(static_cast<float>(SQR(100. - params_Ldetail) + 50.*(100. - params_Ldetail)) * TS * 0.5f);
array2D<float> tilemask_in(TS, TS);
array2D<float> tilemask_out(TS, TS);
if (denoiseLuminance) {
const int border = MAX(2, TS / 16);
for (int i = 0; i < TS; ++i) {
float i1 = abs((i > TS / 2 ? i - TS + 1 : i));
float vmask = (i1 < border ? SQR(sin((rtengine::RT_PI * i1) / (2 * border))) : 1.0f);
float vmask2 = (i1 < 2 * border ? SQR(sin((rtengine::RT_PI * i1) / (2 * border))) : 1.0f);
for (int j = 0; j < TS; ++j) {
float j1 = abs((j > TS / 2 ? j - TS + 1 : j));
tilemask_in[i][j] = (vmask * (j1 < border ? SQR(sin((rtengine::RT_PI * j1) / (2 * border))) : 1.0f)) + epsilon;
tilemask_out[i][j] = (vmask2 * (j1 < 2 * border ? SQR(sin((rtengine::RT_PI * j1) / (2 * border))) : 1.0f)) + epsilon;
}
}
}
int tilesize = 0;
int overlap = 0;
if (settings->leveldnti == 0) {
tilesize = 1024;
overlap = 128;
}
if (settings->leveldnti == 1) {
tilesize = 768;
overlap = 96;
}
int numTries = 0;
if (ponder) {
printf("Tiled denoise processing caused by Automatic Multizone mode\n");
}
bool memoryAllocationFailed = false;
do {
++numTries;
if (numTries == 2) {
printf("1st denoise pass failed due to insufficient memory, starting 2nd (tiled) pass now...\n");
}
int numtiles_W, numtiles_H, tilewidth, tileheight, tileWskip, tileHskip;
Tile_calc(tilesize, overlap, (options.rgbDenoiseThreadLimit == 0 && !ponder) ? (numTries == 1 ? 0 : 2) : 2, imwidth, imheight, numtiles_W, numtiles_H, tilewidth, tileheight, tileWskip, tileHskip);
memoryAllocationFailed = false;
const int numtiles = numtiles_W * numtiles_H;
//output buffer
Imagefloat * dsttmp;
if (numtiles == 1) {
dsttmp = dst;
} else {
dsttmp = new Imagefloat(imwidth, imheight);
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < imheight; ++i) {
for (int j = 0; j < imwidth; ++j) {
dsttmp->r(i, j) = 0.f;
dsttmp->g(i, j) = 0.f;
dsttmp->b(i, j) = 0.f;
}
}
}
//now we have tile dimensions, overlaps
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// According to FFTW-Doc 'it is safe to execute the same plan in parallel by multiple threads', so we now create 4 plans
// outside the parallel region and use them inside the parallel region.
// calculate max size of numblox_W.
int max_numblox_W = ceil((static_cast<float>(MIN(imwidth, tilewidth))) / (offset)) + 2 * blkrad;
// calculate min size of numblox_W.
int min_numblox_W = ceil((static_cast<float>((MIN(imwidth, ((numtiles_W - 1) * tileWskip) + tilewidth)) - ((numtiles_W - 1) * tileWskip))) / (offset)) + 2 * blkrad;
// these are needed only for creation of the plans and will be freed before entering the parallel loop
fftwf_plan plan_forward_blox[2];
fftwf_plan plan_backward_blox[2];
if (denoiseLuminance) {
float *Lbloxtmp = reinterpret_cast<float*>(fftwf_malloc(max_numblox_W * TS * TS * sizeof(float)));
float *fLbloxtmp = reinterpret_cast<float*>(fftwf_malloc(max_numblox_W * TS * TS * sizeof(float)));
int nfwd[2] = {TS, TS};
//for DCT:
fftw_r2r_kind fwdkind[2] = {FFTW_REDFT10, FFTW_REDFT10};
fftw_r2r_kind bwdkind[2] = {FFTW_REDFT01, FFTW_REDFT01};
// Creating the plans with FFTW_MEASURE instead of FFTW_ESTIMATE speeds up the execute a bit
plan_forward_blox[0] = fftwf_plan_many_r2r(2, nfwd, max_numblox_W, Lbloxtmp, nullptr, 1, TS * TS, fLbloxtmp, nullptr, 1, TS * TS, fwdkind, FFTW_MEASURE | FFTW_DESTROY_INPUT);
plan_backward_blox[0] = fftwf_plan_many_r2r(2, nfwd, max_numblox_W, fLbloxtmp, nullptr, 1, TS * TS, Lbloxtmp, nullptr, 1, TS * TS, bwdkind, FFTW_MEASURE | FFTW_DESTROY_INPUT);
plan_forward_blox[1] = fftwf_plan_many_r2r(2, nfwd, min_numblox_W, Lbloxtmp, nullptr, 1, TS * TS, fLbloxtmp, nullptr, 1, TS * TS, fwdkind, FFTW_MEASURE | FFTW_DESTROY_INPUT);
plan_backward_blox[1] = fftwf_plan_many_r2r(2, nfwd, min_numblox_W, fLbloxtmp, nullptr, 1, TS * TS, Lbloxtmp, nullptr, 1, TS * TS, bwdkind, FFTW_MEASURE | FFTW_DESTROY_INPUT);
fftwf_free(Lbloxtmp);
fftwf_free(fLbloxtmp);
}
#ifndef _OPENMP
int numthreads = 1;
#else
// Calculate number of tiles. If less than omp_get_max_threads(), then limit num_threads to number of tiles
int numthreads = MIN(numtiles, omp_get_max_threads());
if (options.rgbDenoiseThreadLimit > 0) {
numthreads = MIN(numthreads, options.rgbDenoiseThreadLimit);
}
#ifdef _OPENMP
denoiseNestedLevels = omp_get_max_threads() / numthreads;
bool oldNested = omp_get_nested();
if (denoiseNestedLevels < 2) {
denoiseNestedLevels = 1;
} else {
omp_set_nested(true);
}
if (options.rgbDenoiseThreadLimit > 0)
while (denoiseNestedLevels * numthreads > options.rgbDenoiseThreadLimit) {
denoiseNestedLevels--;
}
#endif
if (settings->verbose) {
printf("RGB_denoise uses %d main thread(s) and up to %d nested thread(s) for each main thread\n", numthreads, denoiseNestedLevels);
}
#endif
const std::size_t blox_array_size = denoiseNestedLevels * numthreads;
float *LbloxArray[blox_array_size];
float *fLbloxArray[blox_array_size];
for (std::size_t i = 0; i < blox_array_size; ++i) {
LbloxArray[i] = nullptr;
fLbloxArray[i] = nullptr;
}
if (numtiles > 1 && denoiseLuminance) {
for (int i = 0; i < denoiseNestedLevels * numthreads; ++i) {
LbloxArray[i] = reinterpret_cast<float*>(fftwf_malloc(max_numblox_W * TS * TS * sizeof(float)));
fLbloxArray[i] = reinterpret_cast<float*>(fftwf_malloc(max_numblox_W * TS * TS * sizeof(float)));
}
}
TMatrix wiprof = ICCStore::getInstance()->workingSpaceInverseMatrix(params->icm.workingProfile);
//inverse matrix user select
const float wip[3][3] = {
{static_cast<float>(wiprof[0][0]), static_cast<float>(wiprof[0][1]), static_cast<float>(wiprof[0][2])},
{static_cast<float>(wiprof[1][0]), static_cast<float>(wiprof[1][1]), static_cast<float>(wiprof[1][2])},
{static_cast<float>(wiprof[2][0]), static_cast<float>(wiprof[2][1]), static_cast<float>(wiprof[2][2])}
};
TMatrix wprof = ICCStore::getInstance()->workingSpaceMatrix(params->icm.workingProfile);
const float wp[3][3] = {
{static_cast<float>(wprof[0][0]), static_cast<float>(wprof[0][1]), static_cast<float>(wprof[0][2])},
{static_cast<float>(wprof[1][0]), static_cast<float>(wprof[1][1]), static_cast<float>(wprof[1][2])},
{static_cast<float>(wprof[2][0]), static_cast<float>(wprof[2][1]), static_cast<float>(wprof[2][2])}
};
const float wpfast[3][3] = {
{static_cast<float>(wprof[0][0]) / Color::D50x, static_cast<float>(wprof[0][1]) / Color::D50x, static_cast<float>(wprof[0][2]) / Color::D50x},
{static_cast<float>(wprof[1][0]), static_cast<float>(wprof[1][1]), static_cast<float>(wprof[1][2])},
{static_cast<float>(wprof[2][0]) / Color::D50z, static_cast<float>(wprof[2][1]) / Color::D50z, static_cast<float>(wprof[2][2]) / Color::D50z}
};
// begin tile processing of image
#ifdef _OPENMP
#pragma omp parallel num_threads(numthreads) if (numthreads>1)
#endif
{
int pos;
float* noisevarlum;
float* noisevarchrom;
if (numtiles == 1 && isRAW && (useNoiseCCurve || useNoiseLCurve)) {
noisevarlum = lumcalcBuffer;
noisevarchrom = ccalcBuffer;
} else {
noisevarlum = new float[((tileheight + 1) / 2) * ((tilewidth + 1) / 2)];
noisevarchrom = new float[((tileheight + 1) / 2) * ((tilewidth + 1) / 2)];
}
#ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2)
#endif
for (int tiletop = 0; tiletop < imheight; tiletop += tileHskip) {
for (int tileleft = 0; tileleft < imwidth ; tileleft += tileWskip) {
//printf("titop=%d tileft=%d\n",tiletop/tileHskip, tileleft/tileWskip);
pos = (tiletop / tileHskip) * numtiles_W + tileleft / tileWskip ;
int tileright = MIN(imwidth, tileleft + tilewidth);
int tilebottom = MIN(imheight, tiletop + tileheight);
int width = tileright - tileleft;
int height = tilebottom - tiletop;
int width2 = (width + 1) / 2;
float realred, realblue;
float interm_med = static_cast<float>(dnparams.chroma) / 10.0;
float intermred, intermblue;
if (dnparams.redchro > 0.) {
intermred = (dnparams.redchro / 10.);
} else {
intermred = static_cast<float>(dnparams.redchro) / 7.0; //increase slower than linear for more sensit
}
if (dnparams.bluechro > 0.) {
intermblue = (dnparams.bluechro / 10.);
} else {
intermblue = static_cast<float>(dnparams.bluechro) / 7.0; //increase slower than linear for more sensit
}
if (ponder && kall == 2) {
interm_med = ch_M[pos] / 10.f;
intermred = max_r[pos] / 10.f;
intermblue = max_b[pos] / 10.f;
}
if (ponder && kall == 0) {
interm_med = 0.01f;
intermred = 0.f;
intermblue = 0.f;
}
realred = interm_med + intermred;
if (realred <= 0.f) {
realred = 0.001f;
}
realblue = interm_med + intermblue;
if (realblue <= 0.f) {
realblue = 0.001f;
}
const float noisevarab_r = SQR(realred);
const float noisevarab_b = SQR(realblue);
//input L channel
array2D<float> *Lin = nullptr;
//wavelet denoised image
LabImage * labdn = new LabImage(width, height);
//fill tile from image; convert RGB to "luma/chroma"
const float maxNoiseVarab = max(noisevarab_b, noisevarab_r);
if (isRAW) {//image is raw; use channel differences for chroma channels
if (!denoiseMethodRgb) { //lab mode
//modification Jacques feb 2013 and july 2014
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16) num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
for (int i = tiletop; i < tilebottom; ++i) {
const int i1 = i - tiletop;
for (int j = tileleft; j < tileright; ++j) {
const int j1 = j - tileleft;
const float R_ = Color::denoiseIGammaTab[gain * src->r(i, j)];
const float G_ = Color::denoiseIGammaTab[gain * src->g(i, j)];
const float B_ = Color::denoiseIGammaTab[gain * src->b(i, j)];
//apply gamma noise standard (slider)
labdn->L[i1][j1] = R_ < 65535.f ? gamcurve[R_] : Color::gammanf(R_ / 65535.f, gam) * 32768.f;
labdn->a[i1][j1] = G_ < 65535.f ? gamcurve[G_] : Color::gammanf(G_ / 65535.f, gam) * 32768.f;
labdn->b[i1][j1] = B_ < 65535.f ? gamcurve[B_] : Color::gammanf(B_ / 65535.f, gam) * 32768.f;
if (((i1 | j1) & 1) == 0) {
if (numTries == 1) {
noisevarlum[(i1 >> 1) * width2 + (j1 >> 1)] = useNoiseLCurve ? lumcalc[i >> 1][j >> 1] : noisevarL;
noisevarchrom[(i1 >> 1) * width2 + (j1 >> 1)] = useNoiseCCurve ? maxNoiseVarab * ccalc[i >> 1][j >> 1] : 1.f;
} else {
noisevarlum[(i1 >> 1) * width2 + (j1 >> 1)] = lumcalc[i >> 1][j >> 1];
noisevarchrom[(i1 >> 1) * width2 + (j1 >> 1)] = ccalc[i >> 1][j >> 1];
}
}
//end chroma
}
//true conversion xyz=>Lab
Color::RGB2Lab(labdn->L[i1], labdn->a[i1], labdn->b[i1], labdn->L[i1], labdn->a[i1], labdn->b[i1], wpfast, width);
}
} else {//RGB mode
#ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
for (int i = tiletop; i < tilebottom; ++i) {
int i1 = i - tiletop;
for (int j = tileleft; j < tileright; ++j) {
int j1 = j - tileleft;
float X = gain * src->r(i, j);
float Y = gain * src->g(i, j);
float Z = gain * src->b(i, j);
//conversion colorspace to determine luminance with no gamma
X = X < 65535.f ? gamcurve[X] : (Color::gammaf(X / 65535.f, gam, gamthresh, gamslope) * 32768.f);
Y = Y < 65535.f ? gamcurve[Y] : (Color::gammaf(Y / 65535.f, gam, gamthresh, gamslope) * 32768.f);
Z = Z < 65535.f ? gamcurve[Z] : (Color::gammaf(Z / 65535.f, gam, gamthresh, gamslope) * 32768.f);
//end chroma
labdn->L[i1][j1] = Y;
labdn->a[i1][j1] = (X - Y);
labdn->b[i1][j1] = (Y - Z);
if (((i1 | j1) & 1) == 0) {
if (numTries == 1) {
noisevarlum[(i1 >> 1)*width2 + (j1 >> 1)] = useNoiseLCurve ? lumcalc[i >> 1][j >> 1] : noisevarL;
noisevarchrom[(i1 >> 1)*width2 + (j1 >> 1)] = useNoiseCCurve ? maxNoiseVarab * ccalc[i >> 1][j >> 1] : 1.f;
} else {
noisevarlum[(i1 >> 1)*width2 + (j1 >> 1)] = lumcalc[i >> 1][j >> 1];
noisevarchrom[(i1 >> 1)*width2 + (j1 >> 1)] = ccalc[i >> 1][j >> 1];
}
}
}
}
}
} else {//image is not raw; use Lab parametrization
#ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
for (int i = tiletop; i < tilebottom; ++i) {
int i1 = i - tiletop;
for (int j = tileleft; j < tileright; ++j) {
int j1 = j - tileleft;
float L, a, b;
float rLum = src->r(i, j) ; //for denoise curves
float gLum = src->g(i, j) ;
float bLum = src->b(i, j) ;
//use gamma sRGB, not good if TIF (JPG) Output profil not with gamma sRGB (eg : gamma =1.0, or 1.8...)
//very difficult to solve !
// solution ==> save TIF with gamma sRGB and re open
float rtmp = Color::igammatab_srgb[ src->r(i, j) ];
float gtmp = Color::igammatab_srgb[ src->g(i, j) ];
float btmp = Color::igammatab_srgb[ src->b(i, j) ];
//modification Jacques feb 2013
// gamma slider different from raw
rtmp = rtmp < 65535.f ? gamcurve[rtmp] : (Color::gammanf(rtmp / 65535.f, gam) * 32768.f);
gtmp = gtmp < 65535.f ? gamcurve[gtmp] : (Color::gammanf(gtmp / 65535.f, gam) * 32768.f);
btmp = btmp < 65535.f ? gamcurve[btmp] : (Color::gammanf(btmp / 65535.f, gam) * 32768.f);
float X, Y, Z;
Color::rgbxyz(rtmp, gtmp, btmp, X, Y, Z, wp);
//convert Lab
Color::XYZ2Lab(X, Y, Z, L, a, b);
labdn->L[i1][j1] = L;
labdn->a[i1][j1] = a;
labdn->b[i1][j1] = b;
if (((i1 | j1) & 1) == 0) {
float Llum, alum, blum;
if (useNoiseLCurve || useNoiseCCurve) {
float XL, YL, ZL;
Color::rgbxyz(rLum, gLum, bLum, XL, YL, ZL, wp);
Color::XYZ2Lab(XL, YL, ZL, Llum, alum, blum);
}
if (useNoiseLCurve) {
float kN = Llum;
float epsi = 0.01f;
if (kN < 2.f) {
kN = 2.f;
}
if (kN > 32768.f) {
kN = 32768.f;
}
float kinterm = epsi + noiseLCurve[xdivf(kN, 15) * 500.f];
float ki = kinterm * 100.f;
ki += noiseluma;
noisevarlum[(i1 >> 1)*width2 + (j1 >> 1)] = SQR((ki / 125.f) * (1.f + ki / 25.f));
} else {
noisevarlum[(i1 >> 1)*width2 + (j1 >> 1)] = noisevarL;
}
if (useNoiseCCurve) {
float aN = alum;
float bN = blum;
float cN = sqrtf(SQR(aN) + SQR(bN));
if (cN < 100.f) {
cN = 100.f; //avoid divided by zero ???
}
float Cinterm = 1.f + ponderCC * 4.f * noiseCCurve[cN / 60.f];
noisevarchrom[(i1 >> 1)*width2 + (j1 >> 1)] = maxNoiseVarab * SQR(Cinterm);
} else {
noisevarchrom[(i1 >> 1)*width2 + (j1 >> 1)] = 1.f;
}
}
}
}
}
//now perform basic wavelet denoise
//arguments 4 and 5 of wavelet decomposition are max number of wavelet decomposition levels;
//and whether to subsample the image after wavelet filtering. Subsampling is coded as
//binary 1 or 0 for each level, eg subsampling = 0 means no subsampling, 1 means subsample
//the first level only, 7 means subsample the first three levels, etc.
//actual implementation only works with subsampling set to 1
float interm_medT = static_cast<float>(dnparams.chroma) / 10.0;
bool execwavelet = true;
if (!denoiseLuminance && interm_medT < 0.05f && dnparams.median && (dnparams.methodmed == "Lab" || dnparams.methodmed == "Lonly")) {
execwavelet = false; //do not exec wavelet if sliders luminance and chroma are very small and median need
}
//we considered user don't want wavelet
if (settings->leveldnautsimpl == 1 && dnparams.Cmethod != "MAN") {
execwavelet = true;
}
if (settings->leveldnautsimpl == 0 && dnparams.C2method != "MANU") {
execwavelet = true;
}
if (execwavelet) {//gain time if user choose only median sliders L <=1 slider chrom master < 1
wavelet_decomposition* Ldecomp;
wavelet_decomposition* adecomp;
int levwav = 5;
float maxreal = max(realred, realblue);
//increase the level of wavelet if user increase much or very much sliders
if (maxreal < 8.f) {
levwav = 5;
} else if (maxreal < 10.f) {
levwav = 6;
} else if (maxreal < 15.f) {
levwav = 7;
} else {
levwav = 8; //maximum ==> I have increase Maxlevel in cplx_wavelet_dec.h from 8 to 9
}
if (nrQuality == QUALITY_HIGH) {
levwav += settings->nrwavlevel; //increase level for enhanced mode
}
if (levwav > 8) {
levwav = 8;
}
int minsizetile = min(tilewidth, tileheight);
int maxlev2 = 8;
if (minsizetile < 256) {
maxlev2 = 7;
}
if (minsizetile < 128) {
maxlev2 = 6;
}
if (minsizetile < 64) {
maxlev2 = 5;
}
levwav = min(maxlev2, levwav);
// if (settings->verbose) printf("levwavelet=%i noisevarA=%f noisevarB=%f \n",levwav, noisevarab_r, noisevarab_b);
Ldecomp = new wavelet_decomposition(labdn->L[0], labdn->W, labdn->H, levwav, 1, 1, max(1, denoiseNestedLevels));
if (Ldecomp->memoryAllocationFailed) {
memoryAllocationFailed = true;
}
float madL[8][3];
if (!memoryAllocationFailed) {
// precalculate madL, because it's used in adecomp and bdecomp
int maxlvl = Ldecomp->maxlevel();
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) collapse(2) num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
for (int lvl = 0; lvl < maxlvl; ++lvl) {
for (int dir = 1; dir < 4; ++dir) {
// compute median absolute deviation (MAD) of detail coefficients as robust noise estimator
int Wlvl_L = Ldecomp->level_W(lvl);
int Hlvl_L = Ldecomp->level_H(lvl);
float ** WavCoeffs_L = Ldecomp->level_coeffs(lvl);
if (!denoiseMethodRgb) {
madL[lvl][dir - 1] = SQR(Mad(WavCoeffs_L[dir], Wlvl_L * Hlvl_L));
} else {
madL[lvl][dir - 1] = SQR(MadRgb(WavCoeffs_L[dir], Wlvl_L * Hlvl_L));
}
}
}
}
float chresid = 0.f;
float chresidtemp = 0.f;
float chmaxresid = 0.f;
float chmaxresidtemp = 0.f;
adecomp = new wavelet_decomposition(labdn->a[0], labdn->W, labdn->H, levwav, 1, 1, max(1, denoiseNestedLevels));
if (adecomp->memoryAllocationFailed) {
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
if (nrQuality == QUALITY_STANDARD) {
if (!WaveletDenoiseAllAB(*Ldecomp, *adecomp, noisevarchrom, madL, nullptr, 0, noisevarab_r, useNoiseCCurve, autoch, denoiseMethodRgb, denoiseNestedLevels)) { //enhance mode
memoryAllocationFailed = true;
}
} else { /*if (nrQuality==QUALITY_HIGH)*/
if (!WaveletDenoiseAll_BiShrinkAB(*Ldecomp, *adecomp, noisevarchrom, madL, nullptr, 0, noisevarab_r, useNoiseCCurve, autoch, denoiseMethodRgb, denoiseNestedLevels)) { //enhance mode
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
if (!WaveletDenoiseAllAB(*Ldecomp, *adecomp, noisevarchrom, madL, nullptr, 0, noisevarab_r, useNoiseCCurve, autoch, denoiseMethodRgb, denoiseNestedLevels)) {
memoryAllocationFailed = true;
}
}
}
}
if (!memoryAllocationFailed) {
if (kall == 0) {
Noise_residualAB(*adecomp, chresid, chmaxresid, denoiseMethodRgb);
chresidtemp = chresid;
chmaxresidtemp = chmaxresid;
}
adecomp->reconstruct(labdn->a[0]);
}
delete adecomp;
if (!memoryAllocationFailed) {
wavelet_decomposition* bdecomp = new wavelet_decomposition(labdn->b[0], labdn->W, labdn->H, levwav, 1, 1, max(1, denoiseNestedLevels));
if (bdecomp->memoryAllocationFailed) {
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
if (nrQuality == QUALITY_STANDARD) {
if (!WaveletDenoiseAllAB(*Ldecomp, *bdecomp, noisevarchrom, madL, nullptr, 0, noisevarab_b, useNoiseCCurve, autoch, denoiseMethodRgb, denoiseNestedLevels)) { //enhance mode
memoryAllocationFailed = true;
}
} else { /*if (nrQuality==QUALITY_HIGH)*/
if (!WaveletDenoiseAll_BiShrinkAB(*Ldecomp, *bdecomp, noisevarchrom, madL, nullptr, 0, noisevarab_b, useNoiseCCurve, autoch, denoiseMethodRgb, denoiseNestedLevels)) { //enhance mode
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
if (!WaveletDenoiseAllAB(*Ldecomp, *bdecomp, noisevarchrom, madL, nullptr, 0, noisevarab_b, useNoiseCCurve, autoch, denoiseMethodRgb, denoiseNestedLevels)) {
memoryAllocationFailed = true;
}
}
}
}
if (!memoryAllocationFailed) {
if (kall == 0) {
Noise_residualAB(*bdecomp, chresid, chmaxresid, denoiseMethodRgb);
chresid += chresidtemp;
chmaxresid += chmaxresidtemp;
chresid = sqrt(chresid / (6 * (levwav)));
highresi = chresid + 0.66f * (sqrt(chmaxresid) - chresid); //evaluate sigma
nresi = chresid;
}
bdecomp->reconstruct(labdn->b[0]);
}
delete bdecomp;
if (!memoryAllocationFailed) {
if (denoiseLuminance) {
int edge = 0;
if (nrQuality == QUALITY_STANDARD) {
if (!WaveletDenoiseAllL(*Ldecomp, noisevarlum, madL, nullptr, edge, denoiseNestedLevels)) { //enhance mode
memoryAllocationFailed = true;
}
} else { /*if (nrQuality==QUALITY_HIGH)*/
if (!WaveletDenoiseAll_BiShrinkL(*Ldecomp, noisevarlum, madL, nullptr, edge, denoiseNestedLevels)) { //enhance mode
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
if (!WaveletDenoiseAllL(*Ldecomp, noisevarlum, madL, nullptr, edge, denoiseNestedLevels)) {
memoryAllocationFailed = true;
}
}
}
if (!memoryAllocationFailed) {
// copy labdn->L to Lin before it gets modified by reconstruction
Lin = new array2D<float>(width, height);
#ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
for (int i = 0; i < height; ++i) {
for (int j = 0; j < width; ++j) {
(*Lin)[i][j] = labdn->L[i][j];
}
}
Ldecomp->reconstruct(labdn->L[0]);
}
}
}
}
delete Ldecomp;
}
if (!memoryAllocationFailed) {
//wavelet denoised L channel
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// now do detail recovery using block DCT to detect
// patterns missed by wavelet denoise
// blocks are not the same thing as tiles!
// calculation for detail recovery blocks
const int numblox_W = ceil((static_cast<float>(width)) / (offset)) + 2 * blkrad;
const int numblox_H = ceil((static_cast<float>(height)) / (offset)) + 2 * blkrad;
// end of tiling calc
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// Main detail recovery algorithm: Block loop
//DCT block data storage
if (denoiseLuminance /*&& execwavelet*/) {
//residual between input and denoised L channel
array2D<float> Ldetail(width, height, ARRAY2D_CLEAR_DATA);
//pixel weight
array2D<float> totwt(width, height, ARRAY2D_CLEAR_DATA); //weight for combining DCT blocks
if (numtiles == 1) {
for (int i = 0; i < denoiseNestedLevels * numthreads; ++i) {
LbloxArray[i] = reinterpret_cast<float*>(fftwf_malloc(max_numblox_W * TS * TS * sizeof(float)));
fLbloxArray[i] = reinterpret_cast<float*>(fftwf_malloc(max_numblox_W * TS * TS * sizeof(float)));
}
}
#ifdef _OPENMP
int masterThread = omp_get_thread_num();
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
{
#ifdef _OPENMP
int subThread = masterThread * denoiseNestedLevels + omp_get_thread_num();
#else
int subThread = 0;
#endif
// float blurbuffer[TS * TS] ALIGNED64;
float *Lblox = LbloxArray[subThread];
float *fLblox = fLbloxArray[subThread];
float pBuf[width + TS + 2 * blkrad * offset] ALIGNED16;
// float nbrwt[TS * TS] ALIGNED64;
#ifdef _OPENMP
#pragma omp for
#endif
for (int vblk = 0; vblk < numblox_H; ++vblk) {
int top = (vblk - blkrad) * offset;
float * datarow = pBuf + blkrad * offset;
for (int i = 0; i < TS; ++i) {
int row = top + i;
int rr = row;
if (row < 0) {
rr = MIN(-row, height - 1);
} else if (row >= height) {
rr = MAX(0, 2 * height - 2 - row);
}
for (int j = 0; j < labdn->W; ++j) {
datarow[j] = ((*Lin)[rr][j] - labdn->L[rr][j]);
}
for (int j = -blkrad * offset; j < 0; ++j) {
datarow[j] = datarow[MIN(-j, width - 1)];
}
for (int j = width; j < width + TS + blkrad * offset; ++j) {
datarow[j] = datarow[MAX(0, 2 * width - 2 - j)];
}//now we have a padded data row
//now fill this row of the blocks with Lab high pass data
for (int hblk = 0; hblk < numblox_W; ++hblk) {
int left = (hblk - blkrad) * offset;
int indx = (hblk) * TS; //index of block in malloc
if (top + i >= 0 && top + i < height) {
int j;
for (j = 0; j < min((-left), TS); ++j) {
Lblox[(indx + i)*TS + j] = tilemask_in[i][j] * datarow[left + j]; // luma data
}
for (; j < min(TS, width - left); ++j) {
Lblox[(indx + i)*TS + j] = tilemask_in[i][j] * datarow[left + j]; // luma data
totwt[top + i][left + j] += tilemask_in[i][j] * tilemask_out[i][j];
}
for (; j < TS; ++j) {
Lblox[(indx + i)*TS + j] = tilemask_in[i][j] * datarow[left + j]; // luma data
}
} else {
for (int j = 0; j < TS; ++j) {
Lblox[(indx + i)*TS + j] = tilemask_in[i][j] * datarow[left + j]; // luma data
}
}
}
}//end of filling block row
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//fftwf_print_plan (plan_forward_blox);
if (numblox_W == max_numblox_W) {
fftwf_execute_r2r(plan_forward_blox[0], Lblox, fLblox); // DCT an entire row of tiles
} else {
fftwf_execute_r2r(plan_forward_blox[1], Lblox, fLblox); // DCT an entire row of tiles
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// now process the vblk row of blocks for noise reduction
for (int hblk = 0; hblk < numblox_W; ++hblk) {
RGBtile_denoise(fLblox, hblk, noisevar_Ldetail);
// RGBtile_denoise(fLblox, hblk, noisevar_Ldetail, nbrwt, blurbuffer);
}//end of horizontal block loop
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//now perform inverse FT of an entire row of blocks
if (numblox_W == max_numblox_W) {
fftwf_execute_r2r(plan_backward_blox[0], fLblox, Lblox); //for DCT
} else {
fftwf_execute_r2r(plan_backward_blox[1], fLblox, Lblox); //for DCT
}
int topproc = (vblk - blkrad) * offset;
//add row of blocks to output image tile
RGBoutput_tile_row(Lblox, Ldetail, tilemask_out, height, width, topproc);
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
}//end of vertical block loop
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
for (int i = 0; i < height; ++i) {
for (int j = 0; j < width; ++j) {
//may want to include masking threshold for large hipass data to preserve edges/detail
labdn->L[i][j] += Ldetail[i][j] / totwt[i][j]; //note that labdn initially stores the denoised hipass data
}
}
}
if ((metchoice == 1 || metchoice == 2 || metchoice == 3 || metchoice == 4) && dnparams.median) {
float** tmL;
int wid = labdn->W;
int hei = labdn->H;
tmL = new float*[hei];
for (int i = 0; i < hei; ++i) {
tmL[i] = new float[wid];
}
Median medianTypeL = Median::TYPE_3X3_SOFT;
Median medianTypeAB = Median::TYPE_3X3_SOFT;
if (dnparams.medmethod == "soft") {
if (metchoice != 4) {
medianTypeL = medianTypeAB = Median::TYPE_3X3_SOFT;
} else {
medianTypeL = Median::TYPE_3X3_SOFT;
medianTypeAB = Median::TYPE_3X3_SOFT;
}
} else if (dnparams.medmethod == "33") {
if (metchoice != 4) {
medianTypeL = medianTypeAB = Median::TYPE_3X3_STRONG;
} else {
medianTypeL = Median::TYPE_3X3_SOFT;
medianTypeAB = Median::TYPE_3X3_STRONG;
}
} else if (dnparams.medmethod == "55soft") {
if (metchoice != 4) {
medianTypeL = medianTypeAB = Median::TYPE_5X5_SOFT;
} else {
medianTypeL = Median::TYPE_3X3_SOFT;
medianTypeAB = Median::TYPE_5X5_SOFT;
}
} else if (dnparams.medmethod == "55") {
if (metchoice != 4) {
medianTypeL = medianTypeAB = Median::TYPE_5X5_STRONG;
} else {
medianTypeL = Median::TYPE_3X3_STRONG;
medianTypeAB = Median::TYPE_5X5_STRONG;
}
} else if (dnparams.medmethod == "77") {
if (metchoice != 4) {
medianTypeL = medianTypeAB = Median::TYPE_7X7;
} else {
medianTypeL = Median::TYPE_3X3_STRONG;
medianTypeAB = Median::TYPE_7X7;
}
} else if (dnparams.medmethod == "99") {
if (metchoice != 4) {
medianTypeL = medianTypeAB = Median::TYPE_9X9;
} else {
medianTypeL = Median::TYPE_5X5_SOFT;
medianTypeAB = Median::TYPE_9X9;
}
}
if (metchoice == 1 || metchoice == 2 || metchoice == 4) {
Median_Denoise(labdn->L, labdn->L, wid, hei, medianTypeL, dnparams.passes, denoiseNestedLevels, tmL);
}
if (metchoice == 2 || metchoice == 3 || metchoice == 4) {
Median_Denoise(labdn->a, labdn->a, wid, hei, medianTypeAB, dnparams.passes, denoiseNestedLevels, tmL);
Median_Denoise(labdn->b, labdn->b, wid, hei, medianTypeAB, dnparams.passes, denoiseNestedLevels, tmL);
}
for (int i = 0; i < hei; ++i) {
delete[] tmL[i];
}
delete[] tmL;
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// transform denoised "Lab" to output RGB
//calculate mask for feathering output tile overlaps
float Vmask[height + 1] ALIGNED16;
float Hmask[width + 1] ALIGNED16;
float newGain;
if (numtiles > 1) {
for (int i = 0; i < height; ++i) {
Vmask[i] = 1;
}
newGain = 1.f;
if (isRAW) {
newGain = gain;
}
for (int j = 0; j < width; ++j) {
Hmask[j] = 1.f / newGain;
}
for (int i = 0; i < overlap; ++i) {
float mask = SQR(xsinf((rtengine::RT_PI * i) / (2 * overlap)));
if (tiletop > 0) {
Vmask[i] = mask;
}
if (tilebottom < imheight) {
Vmask[height - i] = mask;
}
if (tileleft > 0) {
Hmask[i] = mask / newGain;
}
if (tileright < imwidth) {
Hmask[width - i] = mask / newGain;
}
}
} else {
newGain = isRAW ? 1.f / gain : 1.f;;
}
//convert back to RGB and write to destination array
if (isRAW) {
if (!denoiseMethodRgb) {//Lab mode
realred /= 100.f;
realblue /= 100.f;
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16) num_threads(denoiseNestedLevels)
#endif
for (int i = tiletop; i < tilebottom; ++i) {
int i1 = i - tiletop;
//true conversion Lab==>xyz
Color::Lab2RGBLimit(labdn->L[i1], labdn->a[i1], labdn->b[i1], labdn->L[i1], labdn->a[i1], labdn->b[i1], wip, 9000000.f, 1.f + qhighFactor * realred, 1.f + qhighFactor * realblue, width);
for (int j = tileleft; j < tileright; ++j) {
int j1 = j - tileleft;
float r_ = labdn->L[i1][j1];
float g_ = labdn->a[i1][j1];
float b_ = labdn->b[i1][j1];
//inverse gamma standard (slider)
r_ = r_ < 32768.f ? igamcurve[r_] : (Color::gammanf(r_ / 32768.f, igam) * 65535.f);
g_ = g_ < 32768.f ? igamcurve[g_] : (Color::gammanf(g_ / 32768.f, igam) * 65535.f);
b_ = b_ < 32768.f ? igamcurve[b_] : (Color::gammanf(b_ / 32768.f, igam) * 65535.f);
//readapt arbitrary gamma (inverse from beginning)
r_ = Color::denoiseGammaTab[r_];
g_ = Color::denoiseGammaTab[g_];
b_ = Color::denoiseGammaTab[b_];
if (numtiles == 1) {
dsttmp->r(i, j) = newGain * r_;
dsttmp->g(i, j) = newGain * g_;
dsttmp->b(i, j) = newGain * b_;
} else {
float factor = Vmask[i1] * Hmask[j1];
dsttmp->r(i, j) += factor * r_;
dsttmp->g(i, j) += factor * g_;
dsttmp->b(i, j) += factor * b_;
}
}
}
} else {//RGB mode
#ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels)
#endif
for (int i = tiletop; i < tilebottom; ++i) {
int i1 = i - tiletop;
for (int j = tileleft; j < tileright; ++j) {
int j1 = j - tileleft;
float c_h = sqrt(SQR(labdn->a[i1][j1]) + SQR(labdn->b[i1][j1]));
if (c_h > 3000.f) {
labdn->a[i1][j1] *= 1.f + qhighFactor * realred / 100.f;
labdn->b[i1][j1] *= 1.f + qhighFactor * realblue / 100.f;
}
float Y = labdn->L[i1][j1];
float X = (labdn->a[i1][j1]) + Y;
float Z = Y - (labdn->b[i1][j1]);
X = X < 32768.f ? igamcurve[X] : (Color::gammaf(X / 32768.f, igam, igamthresh, igamslope) * 65535.f);
Y = Y < 32768.f ? igamcurve[Y] : (Color::gammaf(Y / 32768.f, igam, igamthresh, igamslope) * 65535.f);
Z = Z < 32768.f ? igamcurve[Z] : (Color::gammaf(Z / 32768.f, igam, igamthresh, igamslope) * 65535.f);
if (numtiles == 1) {
dsttmp->r(i, j) = newGain * X;
dsttmp->g(i, j) = newGain * Y;
dsttmp->b(i, j) = newGain * Z;
} else {
float factor = Vmask[i1] * Hmask[j1];
dsttmp->r(i, j) += factor * X;
dsttmp->g(i, j) += factor * Y;
dsttmp->b(i, j) += factor * Z;
}
}
}
}
} else {
#ifdef _OPENMP
#pragma omp parallel for num_threads(denoiseNestedLevels)
#endif
for (int i = tiletop; i < tilebottom; ++i) {
int i1 = i - tiletop;
for (int j = tileleft; j < tileright; ++j) {
int j1 = j - tileleft;
//modification Jacques feb 2013
float L = labdn->L[i1][j1];
float a = labdn->a[i1][j1];
float b = labdn->b[i1][j1];
float c_h = sqrt(SQR(a) + SQR(b));
if (c_h > 3000.f) {
a *= 1.f + qhighFactor * realred / 100.f;
b *= 1.f + qhighFactor * realblue / 100.f;
}
float X, Y, Z;
Color::Lab2XYZ(L, a, b, X, Y, Z);
float r_, g_, b_;
Color::xyz2rgb(X, Y, Z, r_, g_, b_, wip);
//gamma slider is different from Raw
r_ = r_ < 32768.f ? igamcurve[r_] : (Color::gammanf(r_ / 32768.f, igam) * 65535.f);
g_ = g_ < 32768.f ? igamcurve[g_] : (Color::gammanf(g_ / 32768.f, igam) * 65535.f);
b_ = b_ < 32768.f ? igamcurve[b_] : (Color::gammanf(b_ / 32768.f, igam) * 65535.f);
if (numtiles == 1) {
dsttmp->r(i, j) = newGain * r_;
dsttmp->g(i, j) = newGain * g_;
dsttmp->b(i, j) = newGain * b_;
} else {
float factor = Vmask[i1] * Hmask[j1];
dsttmp->r(i, j) += factor * r_;
dsttmp->g(i, j) += factor * g_;
dsttmp->b(i, j) += factor * b_;
}
}
}
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
}
delete labdn;
delete Lin;
}//end of tile row
}//end of tile loop
if (numtiles > 1 || !isRAW || (!useNoiseCCurve && !useNoiseLCurve)) {
delete[] noisevarlum;
delete[] noisevarchrom;
}
}
for (size_t i = 0; i < blox_array_size; ++i) {
if (LbloxArray[i]) {
fftwf_free(LbloxArray[i]);
}
if (fLbloxArray[i]) {
fftwf_free(fLbloxArray[i]);
}
}
#ifdef _OPENMP
omp_set_nested(oldNested);
#endif
//copy denoised image to output
if (numtiles > 1) {
if (!memoryAllocationFailed) {
dsttmp->copyData(dst);
} else if (dst != src) {
src->copyData(dst);
}
delete dsttmp;
}
if (!isRAW && !memoryAllocationFailed) {//restore original image gamma
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < dst->getHeight(); ++i) {
for (int j = 0; j < dst->getWidth(); ++j) {
dst->r(i, j) = Color::gammatab_srgb[ dst->r(i, j) ];
dst->g(i, j) = Color::gammatab_srgb[ dst->g(i, j) ];
dst->b(i, j) = Color::gammatab_srgb[ dst->b(i, j) ];
}
}
}
if (denoiseLuminance) {
// destroy the plans
fftwf_destroy_plan(plan_forward_blox[0]);
fftwf_destroy_plan(plan_backward_blox[0]);
fftwf_destroy_plan(plan_forward_blox[1]);
fftwf_destroy_plan(plan_backward_blox[1]);
}
} while (memoryAllocationFailed && numTries < 2 && (options.rgbDenoiseThreadLimit == 0) && !ponder);
if (memoryAllocationFailed) {
printf("tiled denoise failed due to isufficient memory. Output is not denoised!\n");
}
}
//median 3x3 in complement on RGB
if (dnparams.methodmed == "RGB" && dnparams.median) {
//printf("RGB den\n");
int wid = dst->getWidth(), hei = dst->getHeight();
float** tm;
tm = new float*[hei];
for (int i = 0; i < hei; ++i) {
tm[i] = new float[wid];
}
Imagefloat *source;
if (dnparams.luma == 0 && dnparams.chroma == 0) {
source = dst;
} else {
source = src;
}
int methmed = 0;
int border = 1;
if (dnparams.rgbmethod == "soft") {
methmed = 0;
} else if (dnparams.rgbmethod == "33") {
methmed = 1;
} else if (dnparams.rgbmethod == "55") {
methmed = 3;
border = 2;
} else if (dnparams.rgbmethod == "55soft") {
methmed = 2;
border = 2;
}
for (int iteration = 1; iteration <= dnparams.passes; ++iteration) {
#ifdef _OPENMP
#pragma omp parallel
#endif
{
if (methmed < 2)
{
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = 1; i < hei - 1; ++i) {
if (methmed == 0) {
for (int j = 1; j < wid - 1; ++j) {
tm[i][j] = median(source->r(i, j), source->r(i - 1, j), source->r(i + 1, j), source->r(i, j + 1), source->r(i, j - 1)); //3x3 soft
}
} else {
for (int j = 1; j < wid - 1; ++j) {
tm[i][j] = median(source->r(i, j), source->r(i - 1, j), source->r(i + 1, j), source->r(i, j + 1), source->r(i, j - 1), source->r(i - 1, j - 1), source->r(i - 1, j + 1), source->r(i + 1, j - 1), source->r(i + 1, j + 1)); //3x3
}
}
}
} else
{
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = 2; i < hei - 2; ++i) {
if (methmed == 3) {
for (int j = 2; j < wid - 2; ++j) {
tm[i][j] = median(source->r(i, j), source->r(i - 1, j), source->r(i + 1, j), source->r(i, j + 1), source->r(i, j - 1), source->r(i - 1, j - 1), source->r(i - 1, j + 1), source->r(i + 1, j - 1), source->r(i + 1, j + 1),
source->r(i - 2, j), source->r(i + 2, j), source->r(i, j + 2), source->r(i, j - 2), source->r(i - 2, j - 2), source->r(i - 2, j + 2), source->r(i + 2, j - 2), source->r(i + 2, j + 2),
source->r(i - 2, j + 1), source->r(i + 2, j + 1), source->r(i - 1, j + 2), source->r(i - 1, j - 2), source->r(i - 2, j - 1), source->r(i + 2, j - 1), source->r(i + 1, j + 2), source->r(i + 1, j - 2));//5x5
}
} else {
for (int j = 2; j < wid - 2; ++j) {
tm[i][j] = median(
source->r(i, j),
source->r(i - 1, j),
source->r(i + 1, j),
source->r(i, j + 1),
source->r(i, j - 1),
source->r(i - 1, j - 1),
source->r(i - 1, j + 1),
source->r(i + 1, j - 1),
source->r(i + 1, j + 1),
source->r(i + 2, j),
source->r(i - 2, j),
source->r(i, j + 2),
source->r(i, j - 2)
); // 5x5 soft
}
}
}
}
#ifdef _OPENMP
#pragma omp for nowait
#endif
for (int i = border; i < hei - border; ++i)
{
for (int j = border; j < wid - border; ++j) {
dst->r(i, j) = tm[i][j];
}
}
if (methmed < 2)
{
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = 1; i < hei - 1; ++i) {
if (methmed == 0) {
for (int j = 1; j < wid - 1; ++j) {
tm[i][j] = median(source->b(i, j), source->b(i - 1, j), source->b(i + 1, j), source->b(i, j + 1), source->b(i, j - 1));
}
} else {
for (int j = 1; j < wid - 1; ++j) {
tm[i][j] = median(source->b(i, j), source->b(i - 1, j), source->b(i + 1, j), source->b(i, j + 1), source->b(i, j - 1), source->b(i - 1, j - 1), source->b(i - 1, j + 1), source->b(i + 1, j - 1), source->b(i + 1, j + 1));
}
}
}
} else
{
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = 2; i < hei - 2; ++i) {
if (methmed == 3) {
for (int j = 2; j < wid - 2; ++j) {
tm[i][j] = median(source->b(i, j), source->b(i - 1, j), source->b(i + 1, j), source->b(i, j + 1), source->b(i, j - 1), source->b(i - 1, j - 1), source->b(i - 1, j + 1), source->b(i + 1, j - 1), source->b(i + 1, j + 1),
source->b(i - 2, j), source->b(i + 2, j), source->b(i, j + 2), source->b(i, j - 2), source->b(i - 2, j - 2), source->b(i - 2, j + 2), source->b(i + 2, j - 2), source->b(i + 2, j + 2),
source->b(i - 2, j + 1), source->b(i + 2, j + 1), source->b(i - 1, j + 2), source->b(i - 1, j - 2), source->b(i - 2, j - 1), source->b(i + 2, j - 1), source->b(i + 1, j + 2), source->b(i + 1, j - 2)); // 5x5
}
} else {
for (int j = 2; j < wid - 2; ++j) {
tm[i][j] = median(
source->b(i, j),
source->b(i - 1, j),
source->b(i + 1, j),
source->b(i, j + 1),
source->b(i, j - 1),
source->b(i - 1, j - 1),
source->b(i - 1, j + 1),
source->b(i + 1, j - 1),
source->b(i + 1, j + 1),
source->b(i + 2, j),
source->b(i - 2, j),
source->b(i, j + 2),
source->b(i, j - 2)
); // 5x5 soft
}
}
}
}
#ifdef _OPENMP
#pragma omp for nowait
#endif
for (int i = border; i < hei - border; ++i)
{
for (int j = border; j < wid - border; ++j) {
dst->b(i, j) = tm[i][j];
}
}
if (methmed < 2)
{
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = 1; i < hei - 1; ++i) {
if (methmed == 0) {
for (int j = 1; j < wid - 1; ++j) {
tm[i][j] = median(source->g(i, j), source->g(i - 1, j), source->g(i + 1, j), source->g(i, j + 1), source->g(i, j - 1));
}
} else {
for (int j = 1; j < wid - 1; ++j) {
tm[i][j] = median(source->g(i, j), source->g(i - 1, j), source->g(i + 1, j), source->g(i, j + 1), source->g(i, j - 1), source->g(i - 1, j - 1), source->g(i - 1, j + 1), source->g(i + 1, j - 1), source->g(i + 1, j + 1));
}
}
}
} else
{
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = 2; i < hei - 2; ++i) {
if (methmed == 3) {
for (int j = 2; j < wid - 2; ++j) {
tm[i][j] = median(source->g(i, j), source->g(i - 1, j), source->g(i + 1, j), source->g(i, j + 1), source->g(i, j - 1), source->g(i - 1, j - 1), source->g(i - 1, j + 1), source->g(i + 1, j - 1), source->g(i + 1, j + 1),
source->g(i - 2, j), source->g(i + 2, j), source->g(i, j + 2), source->g(i, j - 2), source->g(i - 2, j - 2), source->g(i - 2, j + 2), source->g(i + 2, j - 2), source->g(i + 2, j + 2),
source->g(i - 2, j + 1), source->g(i + 2, j + 1), source->g(i - 1, j + 2), source->g(i - 1, j - 2), source->g(i - 2, j - 1), source->g(i + 2, j - 1), source->g(i + 1, j + 2), source->g(i + 1, j - 2)); // 5x5
}
} else {
for (int j = 2; j < wid - 2; ++j) {
tm[i][j] = median(
source->g(i, j),
source->g(i - 1, j),
source->g(i + 1, j),
source->g(i, j + 1),
source->g(i, j - 1),
source->g(i - 1, j - 1),
source->g(i - 1, j + 1),
source->g(i + 1, j - 1),
source->g(i + 1, j + 1),
source->g(i + 2, j),
source->g(i - 2, j),
source->g(i, j + 2),
source->g(i, j - 2)
); // 5x5 soft
}
}
}
}
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = border; i < hei - border; ++i)
{
for (int j = border; j < wid - border; ++j) {
dst->g(i, j) = tm[i][j];
}
}
}
}
for (int i = 0; i < hei; ++i) {
delete[] tm[i];
}
delete[] tm;
}
//end median
if (noiseLCurve || useNoiseCCurve) {
delete[] lumcalcBuffer;
delete[] lumcalc;
delete[] ccalcBuffer;
delete[] ccalc;
}
//#ifdef _DEBUG
if (settings->verbose) {
t2e.set();
printf("Denoise performed in %d usec:\n", t2e.etime(t1e));
}
//#endif
}//end of main RGB_denoise
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//void ImProcFunctions::RGBtile_denoise(float * fLblox, int hblproc, float noisevar_Ldetail, float * nbrwt, float * blurbuffer) //for DCT
void ImProcFunctions::RGBtile_denoise(float* fLblox, int hblproc, float noisevar_Ldetail) //for DCT
{
float nbrwt[TS * TS] ALIGNED64;
const int blkstart = hblproc * TS * TS;
boxabsblur(fLblox + blkstart, nbrwt, 3, TS, TS, false); //blur neighbor weights for more robust estimation //for DCT
#ifdef __SSE2__
const vfloat noisevar_Ldetailv = F2V(-1.f / noisevar_Ldetail);
const vfloat onev = F2V(1.f);
for (int n = 0; n < TS * TS; n += 4) { //for DCT
const vfloat tempv = onev - xexpf(SQRV(LVF(nbrwt[n])) * noisevar_Ldetailv);
STVF(fLblox[blkstart + n], LVF(fLblox[blkstart + n]) * tempv);
}//output neighbor averaged result
#else
for (int n = 0; n < TS * TS; ++n) { //for DCT
fLblox[blkstart + n] *= (1 - xexpf(-SQR(nbrwt[n]) / noisevar_Ldetail));
}//output neighbor averaged result
#endif
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
void ImProcFunctions::RGBoutput_tile_row(float *bloxrow_L, float ** Ldetail, float ** tilemask_out, int height, int width, int top)
{
const int numblox_W = ceil((static_cast<float>(width)) / (offset));
const float DCTnorm = 1.0f / (4 * TS * TS); //for DCT
int imin = MAX(0, -top);
int bottom = MIN(top + TS, height);
int imax = bottom - top;
//add row of tiles to output image
for (int i = imin; i < imax; ++i) {
for (int hblk = 0; hblk < numblox_W; ++hblk) {
int left = (hblk - blkrad) * offset;
int right = MIN(left + TS, width);
int jmin = MAX(0, -left);
int jmax = right - left;
int indx = hblk * TS;
for (int j = jmin; j < jmax; ++j) { // this loop gets auto vectorized by gcc
Ldetail[top + i][left + j] += tilemask_out[i][j] * bloxrow_L[(indx + i) * TS + j] * DCTnorm; //for DCT
}
}
}
}
/*
#undef TS
#undef fTS
#undef offset
#undef epsilon
*/
float ImProcFunctions::Mad(const float * DataList, const int datalen)
{
if (datalen <= 1) { // Avoid possible buffer underrun
return 0;
}
//computes Median Absolute Deviation
//DataList values should mostly have abs val < 256 because we are in Lab mode
int histo[256] ALIGNED64 = {0};
//calculate histogram of absolute values of wavelet coeffs
for (int i = 0; i < datalen; ++i) {
histo[static_cast<int>(rtengine::min(255.f, fabsf(DataList[i])))]++;
}
//find median of histogram
int median = 0, count = 0;
while (count < datalen / 2) {
count += histo[median];
++median;
}
int count_ = count - histo[median - 1];
// interpolate
return (((median - 1) + (datalen / 2 - count_) / (static_cast<float>(count - count_))) / 0.6745);
}
float ImProcFunctions::MadRgb(const float * DataList, const int datalen)
{
if (datalen <= 1) { // Avoid possible buffer underrun
return 0;
}
//computes Median Absolute Deviation
//DataList values should mostly have abs val < 65536 because we are in RGB mode
int * histo = new int[65536];
for (int i = 0; i < 65536; ++i) {
histo[i] = 0;
}
//calculate histogram of absolute values of wavelet coeffs
for (int i = 0; i < datalen; ++i) {
histo[static_cast<int>(rtengine::min(65535.f, fabsf(DataList[i])))]++;
}
//find median of histogram
int median = 0, count = 0;
while (count < datalen / 2) {
count += histo[median];
++median;
}
int count_ = count - histo[median - 1];
// interpolate
delete[] histo;
return (((median - 1) + (datalen / 2 - count_) / (static_cast<float>(count - count_))) / 0.6745);
}
void ImProcFunctions::Noise_residualAB(const wavelet_decomposition &WaveletCoeffs_ab, float &chresid, float &chmaxresid, bool denoiseMethodRgb)
{
float resid = 0.f;
float maxresid = 0.f;
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) collapse(2) reduction(+:resid) reduction(max:maxresid) num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
for (int lvl = 0; lvl < WaveletCoeffs_ab.maxlevel(); ++lvl) {
// compute median absolute deviation (MAD) of detail coefficients as robust noise estimator
for (int dir = 1; dir < 4; ++dir) {
const int Wlvl_ab = WaveletCoeffs_ab.level_W(lvl);
const int Hlvl_ab = WaveletCoeffs_ab.level_H(lvl);
float ** WavCoeffs_ab = WaveletCoeffs_ab.level_coeffs(lvl);
const float madC = SQR(denoiseMethodRgb ? MadRgb(WavCoeffs_ab[dir], Wlvl_ab * Hlvl_ab) : Mad(WavCoeffs_ab[dir], Wlvl_ab * Hlvl_ab));
resid += madC;
if (madC > maxresid) {
maxresid = madC;
}
}
}
chresid = resid;
chmaxresid = maxresid;
}
bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(const wavelet_decomposition &WaveletCoeffs_L, float *noisevarlum, float madL[8][3], float * vari, int edge, int denoiseNestedLevels)
{
int maxlvl = min(WaveletCoeffs_L.maxlevel(), 5);
const float eps = 0.01f;
if (edge == 1 || edge == 3) {
maxlvl = 4; //for refine denoise edge wavelet
}
if (edge == 2) {
maxlvl = 7; //for locallab denoise
}
int maxWL = 0, maxHL = 0;
for (int lvl = 0; lvl < maxlvl; ++lvl) {
if (WaveletCoeffs_L.level_W(lvl) > maxWL) {
maxWL = WaveletCoeffs_L.level_W(lvl);
}
if (WaveletCoeffs_L.level_H(lvl) > maxHL) {
maxHL = WaveletCoeffs_L.level_H(lvl);
}
}
bool memoryAllocationFailed = false;
#ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
{
float *buffer[3];
buffer[0] = new (std::nothrow) float[maxWL * maxHL + 32];
buffer[1] = new (std::nothrow) float[maxWL * maxHL + 64];
buffer[2] = new (std::nothrow) float[maxWL * maxHL + 96];
if (buffer[0] == nullptr || buffer[1] == nullptr || buffer[2] == nullptr) {
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
#ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2)
#endif
for (int lvl = maxlvl - 1; lvl >= 0; lvl--) { //for levels less than max, use level diff to make edge mask
for (int dir = 1; dir < 4; ++dir) {
int Wlvl_L = WaveletCoeffs_L.level_W(lvl);
int Hlvl_L = WaveletCoeffs_L.level_H(lvl);
float ** WavCoeffs_L = WaveletCoeffs_L.level_coeffs(lvl);
if (lvl == maxlvl - 1) {
// int edge = 0;
ShrinkAllL(WaveletCoeffs_L, buffer, lvl, dir, noisevarlum, madL[lvl], vari, edge);
} else {
//simple wavelet shrinkage
float * sfave = buffer[0] + 32;
float * sfaved = buffer[2] + 96;
// float * blurBuffer = buffer[2] + 96;
float mad_Lr = madL[lvl][dir - 1];
/*
if ((edge == 1 || edge == 2 || edge == 3) && vari) {
noisevarlum = blurBuffer; // we need one buffer, but fortunately we don't have to allocate a new one because we can use blurBuffer
for (int i = 0; i < Wlvl_L * Hlvl_L; ++i) {
noisevarlum[i] = vari[lvl];
}
}
*/
float *nvl = nullptr;
nvl = new float[Hlvl_L * Wlvl_L];
for (int i = 0; i < Hlvl_L * Wlvl_L; ++i) {
nvl[i] = 0.f;
}
if ((edge == 1 || edge == 2 || edge == 3) && vari) {
// nvl = blurBuffer; // we need one buffer, but fortunately we don't have to allocate a new one because we can use blurBuffer
if ((edge == 1 || edge == 3)) {
for (int i = 0; i < Hlvl_L * Wlvl_L; ++i) {
nvl[i] = vari[lvl];
}
}
if (edge == 2) {
for (int i = 0; i < Hlvl_L * Wlvl_L; ++i) {
nvl[i] = vari[lvl] * SQR(noisevarlum[i]);
}
}
}
else {
for (int i = 0; i < Hlvl_L * Wlvl_L; ++i) {
nvl[i] = noisevarlum[i];
}
}
float levelFactor = mad_Lr * 5.f / (lvl + 1);
#ifdef __SSE2__
vfloat mad_Lv;
vfloat ninev = F2V(9.0f);
vfloat epsv = F2V(eps);
vfloat mag_Lv;
vfloat levelFactorv = F2V(levelFactor);
int coeffloc_L;
for (coeffloc_L = 0; coeffloc_L < Hlvl_L * Wlvl_L - 3; coeffloc_L += 4) {
mad_Lv = LVFU(nvl[coeffloc_L]) * levelFactorv;
mag_Lv = SQRV(LVFU(WavCoeffs_L[dir][coeffloc_L]));
STVFU(sfave[coeffloc_L], mag_Lv / (mag_Lv + mad_Lv * xexpf(-mag_Lv / (mad_Lv * ninev)) + epsv));
}
for (; coeffloc_L < Hlvl_L * Wlvl_L; ++coeffloc_L) {
float mag_L = SQR(WavCoeffs_L[dir][coeffloc_L]);
sfave[coeffloc_L] = mag_L / (mag_L + levelFactor * nvl[coeffloc_L] * xexpf(-mag_L / (9.f * levelFactor * nvl[coeffloc_L])) + eps);
}
#else
for (int i = 0; i < Hlvl_L; ++i) {
for (int j = 0; j < Wlvl_L; ++j) {
int coeffloc_L = i * Wlvl_L + j;
float mag_L = SQR(WavCoeffs_L[dir][coeffloc_L]);
sfave[coeffloc_L] = mag_L / (mag_L + levelFactor * nvl[coeffloc_L] * xexpf(-mag_L / (9.f * levelFactor * nvl[coeffloc_L])) + eps);
}
}
#endif
boxblur(sfave, sfaved, lvl + 2, Wlvl_L, Hlvl_L, false); //increase smoothness by locally averaging shrinkage
// boxblur(sfave, sfaved, blurBuffer, lvl + 2, lvl + 2, Wlvl_L, Hlvl_L); //increase smoothness by locally averaging shrinkage
#ifdef __SSE2__
vfloat sfavev;
vfloat sf_Lv;
for (coeffloc_L = 0; coeffloc_L < Hlvl_L * Wlvl_L - 3; coeffloc_L += 4) {
sfavev = LVFU(sfaved[coeffloc_L]);
sf_Lv = LVFU(sfave[coeffloc_L]);
STVFU(WavCoeffs_L[dir][coeffloc_L], LVFU(WavCoeffs_L[dir][coeffloc_L]) * (SQRV(sfavev) + SQRV(sf_Lv)) / (sfavev + sf_Lv + epsv));
//use smoothed shrinkage unless local shrinkage is much less
}
// few remaining pixels
for (; coeffloc_L < Hlvl_L * Wlvl_L; ++coeffloc_L) {
float sf_L = sfave[coeffloc_L];
//use smoothed shrinkage unless local shrinkage is much less
WavCoeffs_L[dir][coeffloc_L] *= (SQR(sfaved[coeffloc_L]) + SQR(sf_L)) / (sfaved[coeffloc_L] + sf_L + eps);
}//now luminance coeffs are denoised
#else
for (int i = 0; i < Hlvl_L; ++i) {
for (int j = 0; j < Wlvl_L; ++j) {
int coeffloc_L = i * Wlvl_L + j;
float sf_L = sfave[coeffloc_L];
//use smoothed shrinkage unless local shrinkage is much less
WavCoeffs_L[dir][coeffloc_L] *= (SQR(sfaved[coeffloc_L]) + SQR(sf_L)) / (sfaved[coeffloc_L] + sf_L + eps);
}//now luminance coeffs are denoised
}
#endif
delete [] nvl;
}
}
}
}
for (int i = 3; i >= 0; i--) {
if (buffer[i] != nullptr) {
delete[] buffer[i];
}
}
}
return (!memoryAllocationFailed);
}
bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(const wavelet_decomposition &WaveletCoeffs_L, const wavelet_decomposition &WaveletCoeffs_ab, float *noisevarchrom, float madL[8][3], float *variC, int local, float noisevar_ab, const bool useNoiseCCurve, bool autoch, bool denoiseMethodRgb, int denoiseNestedLevels)
{
int maxlvl = WaveletCoeffs_L.maxlevel();
if (local == 2) {
maxlvl = 7; //for local denoise
}
if (local == 3) {
maxlvl = 4; //for shape detection
}
if (autoch && noisevar_ab <= 0.001f) {
noisevar_ab = 0.02f;
}
float madab[8][3];
int maxWL = 0, maxHL = 0;
for (int lvl = 0; lvl < maxlvl; ++lvl) {
if (WaveletCoeffs_L.level_W(lvl) > maxWL) {
maxWL = WaveletCoeffs_L.level_W(lvl);
}
if (WaveletCoeffs_L.level_H(lvl) > maxHL) {
maxHL = WaveletCoeffs_L.level_H(lvl);
}
}
bool memoryAllocationFailed = false;
#ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
{
float *buffer[3];
buffer[0] = new (std::nothrow) float[maxWL * maxHL + 32];
buffer[1] = new (std::nothrow) float[maxWL * maxHL + 64];
buffer[2] = new (std::nothrow) float[maxWL * maxHL + 96];
if (buffer[0] == nullptr || buffer[1] == nullptr || buffer[2] == nullptr) {
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
#ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2)
#endif
for (int lvl = 0; lvl < maxlvl; ++lvl) {
for (int dir = 1; dir < 4; ++dir) {
// compute median absolute deviation (MAD) of detail coefficients as robust noise estimator
int Wlvl_ab = WaveletCoeffs_ab.level_W(lvl);
int Hlvl_ab = WaveletCoeffs_ab.level_H(lvl);
float ** WavCoeffs_ab = WaveletCoeffs_ab.level_coeffs(lvl);
if (!denoiseMethodRgb) {
madab[lvl][dir - 1] = SQR(Mad(WavCoeffs_ab[dir], Wlvl_ab * Hlvl_ab));
} else {
madab[lvl][dir - 1] = SQR(MadRgb(WavCoeffs_ab[dir], Wlvl_ab * Hlvl_ab));
}
}
}
#ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2)
#endif
for (int lvl = maxlvl - 1; lvl >= 0; lvl--) { //for levels less than max, use level diff to make edge mask
for (int dir = 1; dir < 4; ++dir) {
int Wlvl_ab = WaveletCoeffs_ab.level_W(lvl);
int Hlvl_ab = WaveletCoeffs_ab.level_H(lvl);
float ** WavCoeffs_L = WaveletCoeffs_L.level_coeffs(lvl);
float ** WavCoeffs_ab = WaveletCoeffs_ab.level_coeffs(lvl);
if (lvl == maxlvl - 1) {
ShrinkAllAB(WaveletCoeffs_L, WaveletCoeffs_ab, buffer, lvl, dir, noisevarchrom, noisevar_ab, useNoiseCCurve, autoch, denoiseMethodRgb, madL[lvl], nullptr, 0, madab[lvl], true);
} else {
//simple wavelet shrinkage
float noisevarfc;
float mad_Lr = madL[lvl][dir - 1];
float *nvc = nullptr;
nvc = new float[Hlvl_ab * Wlvl_ab];
if ((local == 2 || local == 3) && variC && useNoiseCCurve) {
noisevarfc = variC[lvl];
for (int p = 0; p < Hlvl_ab * Wlvl_ab; p++) {
nvc[p] = 10.f * sqrt(variC[lvl]) * SQR(1.f + 4.f * noisevarchrom[p]);
}
} else {
noisevarfc = noisevar_ab;
for (int p = 0; p < Hlvl_ab * Wlvl_ab; p++) {
nvc[p] = noisevarchrom[p];
}
}
// float mad_abr = useNoiseCCurve ? noisevar_ab * madab[lvl][dir - 1] : SQR(noisevar_ab) * madab[lvl][dir - 1];
float mad_abr = useNoiseCCurve ? noisevarfc * madab[lvl][dir - 1] : SQR(noisevarfc) * madab[lvl][dir - 1];
if (noisevarfc > 0.001f) {
#ifdef __SSE2__
vfloat onev = F2V(1.f);
vfloat mad_abrv = F2V(mad_abr);
vfloat rmad_Lm9v = onev / F2V(mad_Lr * 9.f);
vfloat mad_abv;
vfloat mag_Lv, mag_abv;
vfloat tempabv;
int coeffloc_ab;
for (coeffloc_ab = 0; coeffloc_ab < Hlvl_ab * Wlvl_ab - 3; coeffloc_ab += 4) {
mad_abv = LVFU(nvc[coeffloc_ab]) * mad_abrv;
tempabv = LVFU(WavCoeffs_ab[dir][coeffloc_ab]);
mag_Lv = LVFU(WavCoeffs_L[dir][coeffloc_ab]);
mag_abv = SQRV(tempabv);
mag_Lv = SQRV(mag_Lv) * rmad_Lm9v;
STVFU(WavCoeffs_ab[dir][coeffloc_ab], tempabv * SQRV((onev - xexpf(-(mag_abv / mad_abv) - (mag_Lv)))));
}
// few remaining pixels
for (; coeffloc_ab < Hlvl_ab * Wlvl_ab; ++coeffloc_ab) {
float mag_L = SQR(WavCoeffs_L[dir][coeffloc_ab ]);
float mag_ab = SQR(WavCoeffs_ab[dir][coeffloc_ab]);
WavCoeffs_ab[dir][coeffloc_ab] *= SQR(1.f - xexpf(-(mag_ab / (nvc[coeffloc_ab] * mad_abr)) - (mag_L / (9.f * mad_Lr)))/*satfactor_a*/);
}//now chrominance coefficients are denoised
#else
for (int i = 0; i < Hlvl_ab; ++i) {
for (int j = 0; j < Wlvl_ab; ++j) {
int coeffloc_ab = i * Wlvl_ab + j;
float mag_L = SQR(WavCoeffs_L[dir][coeffloc_ab ]);
float mag_ab = SQR(WavCoeffs_ab[dir][coeffloc_ab]);
WavCoeffs_ab[dir][coeffloc_ab] *= SQR(1.f - xexpf(-(mag_ab / (nvc[coeffloc_ab] * mad_abr)) - (mag_L / (9.f * mad_Lr)))/*satfactor_a*/);
}
}//now chrominance coefficients are denoised
#endif
}
delete [] nvc;
}
}
}
}
for (int i = 2; i >= 0; i--) {
delete[] buffer[i];
}
}
return (!memoryAllocationFailed);
}
bool ImProcFunctions::WaveletDenoiseAllL(const wavelet_decomposition &WaveletCoeffs_L, float *noisevarlum, float madL[8][3], float * vari, int edge, int denoiseNestedLevels)//mod JD
{
int maxlvl = min(WaveletCoeffs_L.maxlevel(), 5);
if (edge == 1 || edge == 3) {
maxlvl = 4; //for refine denoise edge wavelet
}
if (edge == 2) {
maxlvl = 7; //for locallab denoise
}
int maxWL = 0, maxHL = 0;
for (int lvl = 0; lvl < maxlvl; ++lvl) {
if (WaveletCoeffs_L.level_W(lvl) > maxWL) {
maxWL = WaveletCoeffs_L.level_W(lvl);
}
if (WaveletCoeffs_L.level_H(lvl) > maxHL) {
maxHL = WaveletCoeffs_L.level_H(lvl);
}
}
bool memoryAllocationFailed = false;
#ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
{
float *buffer[4];
buffer[0] = new (std::nothrow) float[maxWL * maxHL + 32];
buffer[1] = new (std::nothrow) float[maxWL * maxHL + 64];
buffer[2] = new (std::nothrow) float[maxWL * maxHL + 96];
buffer[3] = new (std::nothrow) float[maxWL * maxHL + 128];
if (buffer[0] == nullptr || buffer[1] == nullptr || buffer[2] == nullptr || buffer[3] == nullptr) {
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
#ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2)
#endif
for (int lvl = 0; lvl < maxlvl; ++lvl) {
for (int dir = 1; dir < 4; ++dir) {
ShrinkAllL(WaveletCoeffs_L, buffer, lvl, dir, noisevarlum, madL[lvl], vari, edge);
}
}
}
for (int i = 3; i >= 0; i--) {
delete[] buffer[i];
}
}
return (!memoryAllocationFailed);
}
bool ImProcFunctions::WaveletDenoiseAllAB(const wavelet_decomposition &WaveletCoeffs_L, const wavelet_decomposition &WaveletCoeffs_ab,
float *noisevarchrom, float madL[8][3], float *variC, int local, float noisevar_ab, const bool useNoiseCCurve, bool autoch, bool denoiseMethodRgb, int denoiseNestedLevels)//mod JD
{
int maxlvl = WaveletCoeffs_L.maxlevel();
if (local == 2) {
maxlvl = 7; //for local denoise
}
if (local == 3) {
maxlvl = 4; //for shape detection
}
int maxWL = 0, maxHL = 0;
for (int lvl = 0; lvl < maxlvl; ++lvl) {
if (WaveletCoeffs_L.level_W(lvl) > maxWL) {
maxWL = WaveletCoeffs_L.level_W(lvl);
}
if (WaveletCoeffs_L.level_H(lvl) > maxHL) {
maxHL = WaveletCoeffs_L.level_H(lvl);
}
}
bool memoryAllocationFailed = false;
#ifdef _OPENMP
#pragma omp parallel num_threads(denoiseNestedLevels) if (denoiseNestedLevels>1)
#endif
{
float *buffer[3];
buffer[0] = new (std::nothrow) float[maxWL * maxHL + 32];
buffer[1] = new (std::nothrow) float[maxWL * maxHL + 64];
buffer[2] = new (std::nothrow) float[maxWL * maxHL + 96];
if (buffer[0] == nullptr || buffer[1] == nullptr || buffer[2] == nullptr) {
memoryAllocationFailed = true;
}
if (!memoryAllocationFailed) {
#ifdef _OPENMP
#pragma omp for schedule(dynamic) collapse(2) nowait
#endif
for (int lvl = 0; lvl < maxlvl; ++lvl) {
for (int dir = 1; dir < 4; ++dir) {
ShrinkAllAB(WaveletCoeffs_L, WaveletCoeffs_ab, buffer, lvl, dir, noisevarchrom, noisevar_ab, useNoiseCCurve, autoch, denoiseMethodRgb, madL[lvl], variC, local, nullptr, 0);
}
}
}
for (int i = 2; i >= 0; i--) {
if (buffer[i] != nullptr) {
delete[] buffer[i];
}
}
}
return (!memoryAllocationFailed);
}
void ImProcFunctions::ShrinkAllL(const wavelet_decomposition &WaveletCoeffs_L, float **buffer, int level, int dir,
float *noisevarlum, float * madL, float * vari, int edge)
{
//simple wavelet shrinkage
const float eps = 0.01f;
float * sfave = buffer[0] + 32;
float * sfaved = buffer[1] + 64;
// float * blurBuffer = buffer[2] + 96;
const int W_L = WaveletCoeffs_L.level_W(level);
const int H_L = WaveletCoeffs_L.level_H(level);
float ** WavCoeffs_L = WaveletCoeffs_L.level_coeffs(level);
const float mad_L = madL[dir - 1] ;
const float levelFactor = mad_L * 5.f / static_cast<float>(level + 1);
float *nvl = nullptr;
nvl = new float[ H_L * W_L];
for (int i = 0; i < W_L * H_L; ++i) {
nvl[i] = 0.f;
}
if ((edge == 1 || edge == 2 || edge == 3) && vari) {
// nvl = blurBuffer; // we need one buffer, but fortunately we don't have to allocate a new one because we can use blurBuffer
if ((edge == 1 || edge == 3)) {
for (int i = 0; i < W_L * H_L; ++i) {
nvl[i] = vari[level]; //* SQR(1.f + 4.f * noisevarchrom[p]);
}
}
if (edge == 2) {
for (int i = 0; i < W_L * H_L; ++i) {
nvl[i] = vari[level] * SQR(noisevarlum[i]);
}
}
}
else {
for (int i = 0; i < W_L * H_L; ++i) {
nvl[i] = noisevarlum[i];
}
}
int i = 0;
#ifdef __SSE2__
const vfloat levelFactorv = F2V(levelFactor);
const vfloat ninev = F2V(9.f);
const vfloat epsv = F2V(eps);
for (i = 0; i < W_L * H_L - 3; i += 4) {
const vfloat mad_Lv = LVFU(noisevarlum[i]) * levelFactorv;
const vfloat magv = SQRV(LVFU(WavCoeffs_L[dir][i]));
STVFU(sfave[i], magv / (magv + mad_Lv * xexpf(-magv / (ninev * mad_Lv)) + epsv));
}
#endif
// few remaining pixels
for (; i < W_L * H_L; ++i) {
float mag = SQR(WavCoeffs_L[dir][i]);
sfave[i] = mag / (mag + levelFactor * nvl[i] * xexpf(-mag / (9 * levelFactor * nvl[i])) + eps);
}
boxblur(sfave, sfaved, level + 2, W_L, H_L, false); //increase smoothness by locally averaging shrinkage
i = 0;
#ifdef __SSE2__
for (; i < W_L * H_L - 3; i += 4) {
const vfloat sfv = LVFU(sfave[i]);
//use smoothed shrinkage unless local shrinkage is much less
STVFU(WavCoeffs_L[dir][i], LVFU(WavCoeffs_L[dir][i]) * (SQRV(LVFU(sfaved[i])) + SQRV(sfv)) / (LVFU(sfaved[i]) + sfv + epsv));
}
#endif
// few remaining pixels
for (; i < W_L * H_L; ++i) {
const float sf = sfave[i];
//use smoothed shrinkage unless local shrinkage is much less
WavCoeffs_L[dir][i] *= (SQR(sfaved[i]) + SQR(sf)) / (sfaved[i] + sf + eps);
}//now luminance coefficients are denoised
delete [] nvl;
}
void ImProcFunctions::ShrinkAllAB(const wavelet_decomposition & WaveletCoeffs_L, const wavelet_decomposition & WaveletCoeffs_ab, float **buffer, int level, int dir,
float * noisevarchrom, float noisevar_ab, const bool useNoiseCCurve, bool autoch,
bool denoiseMethodRgb, float * madL, float * variC, int local, float * madaab, bool madCalculated)
{
//simple wavelet shrinkage
const float eps = 0.01f;
if (autoch && noisevar_ab <= 0.001f) {
noisevar_ab = 0.02f;
}
float * sfaveab = buffer[0] + 32;
float * sfaveabd = buffer[1] + 64;
// float * blurBuffer = buffer[2] + 96;
int W_ab = WaveletCoeffs_ab.level_W(level);
int H_ab = WaveletCoeffs_ab.level_H(level);
float ** WavCoeffs_L = WaveletCoeffs_L.level_coeffs(level);
float ** WavCoeffs_ab = WaveletCoeffs_ab.level_coeffs(level);
float madab;
float mad_L = madL[dir - 1];
if (madCalculated) {
madab = madaab[dir - 1];
} else {
if (!denoiseMethodRgb) {
madab = SQR(Mad(WavCoeffs_ab[dir], W_ab * H_ab));
} else {
madab = SQR(MadRgb(WavCoeffs_ab[dir], W_ab * H_ab));
}
}
float noisevarfc;
float *nvc = nullptr;
nvc = new float[ H_ab * W_ab];
if ((local == 2 || local == 3) && variC && useNoiseCCurve) {
noisevarfc = variC[level];
for (int p = 0; p < H_ab * W_ab; p++) {
nvc[p] = 10.f * sqrt(variC[level]) * SQR(1.f + 4.f * noisevarchrom[p]);
}
} else {
noisevarfc = noisevar_ab;
for (int p = 0; p < H_ab * W_ab; p++) {
nvc[p] = noisevarchrom[p];
}
}
// printf("varfc=%f nvc0=%f nvc1=%f nvc2=%f\n", noisevarfc, nvc[10], nvc[H_ab * W_ab /3], nvc[H_ab * W_ab /2]);
if (noisevarfc > 0.001f) {//noisevar_ab
//madab = useNoiseCCurve ? madab : madab * noisevar_ab;
madab = useNoiseCCurve ? madab : madab * noisevarfc;
#ifdef __SSE2__
vfloat onev = F2V(1.f);
vfloat mad_abrv = F2V(madab);
vfloat rmadLm9v = onev / F2V(mad_L * 9.f);
vfloat mad_abv ;
vfloat mag_Lv, mag_abv;
int coeffloc_ab;
for (coeffloc_ab = 0; coeffloc_ab < H_ab * W_ab - 3; coeffloc_ab += 4) {
mad_abv = LVFU(nvc[coeffloc_ab]) * mad_abrv;
mag_Lv = LVFU(WavCoeffs_L[dir][coeffloc_ab]);
mag_abv = SQRV(LVFU(WavCoeffs_ab[dir][coeffloc_ab]));
mag_Lv = (SQRV(mag_Lv)) * rmadLm9v;
STVFU(sfaveab[coeffloc_ab], (onev - xexpf(-(mag_abv / mad_abv) - (mag_Lv))));
}
// few remaining pixels
for (; coeffloc_ab < H_ab * W_ab; ++coeffloc_ab) {
float mag_L = SQR(WavCoeffs_L[dir][coeffloc_ab]);
float mag_ab = SQR(WavCoeffs_ab[dir][coeffloc_ab]);
sfaveab[coeffloc_ab] = (1.f - xexpf(-(mag_ab / (nvc[coeffloc_ab] * madab)) - (mag_L / (9.f * mad_L))));
}//now chrominance coefficients are denoised
#else
for (int i = 0; i < H_ab; ++i) {
for (int j = 0; j < W_ab; ++j) {
int coeffloc_ab = i * W_ab + j;
float mag_L = SQR(WavCoeffs_L[dir][coeffloc_ab]);
float mag_ab = SQR(WavCoeffs_ab[dir][coeffloc_ab]);
sfaveab[coeffloc_ab] = (1.f - xexpf(-(mag_ab / (nvc[coeffloc_ab] * madab)) - (mag_L / (9.f * mad_L))));
}
}//now chrominance coefficients are denoised
#endif
boxblur(sfaveab, sfaveabd, level + 2, W_ab, H_ab, false); //increase smoothness by locally averaging shrinkage
// boxblur(sfaveab, sfaveabd, blurBuffer, level + 2, level + 2, W_ab, H_ab); //increase smoothness by locally averaging shrinkage
#ifdef __SSE2__
vfloat epsv = F2V(eps);
vfloat sfabv;
vfloat sfaveabv;
for (coeffloc_ab = 0; coeffloc_ab < H_ab * W_ab - 3; coeffloc_ab += 4) {
sfabv = LVFU(sfaveab[coeffloc_ab]);
sfaveabv = LVFU(sfaveabd[coeffloc_ab]);
//use smoothed shrinkage unless local shrinkage is much less
STVFU(WavCoeffs_ab[dir][coeffloc_ab], LVFU(WavCoeffs_ab[dir][coeffloc_ab]) * (SQRV(sfaveabv) + SQRV(sfabv)) / (sfaveabv + sfabv + epsv));
}
// few remaining pixels
for (; coeffloc_ab < H_ab * W_ab; ++coeffloc_ab) {
//modification Jacques feb 2013
float sfab = sfaveab[coeffloc_ab];
//use smoothed shrinkage unless local shrinkage is much less
WavCoeffs_ab[dir][coeffloc_ab] *= (SQR(sfaveabd[coeffloc_ab]) + SQR(sfab)) / (sfaveabd[coeffloc_ab] + sfab + eps);
}//now chrominance coefficients are denoised
#else
for (int i = 0; i < H_ab; ++i) {
for (int j = 0; j < W_ab; ++j) {
int coeffloc_ab = i * W_ab + j;
float sfab = sfaveab[coeffloc_ab];
//use smoothed shrinkage unless local shrinkage is much less
WavCoeffs_ab[dir][coeffloc_ab] *= (SQR(sfaveabd[coeffloc_ab]) + SQR(sfab)) / (sfaveabd[coeffloc_ab] + sfab + eps);
}//now chrominance coefficients are denoised
}
#endif
}
delete [] nvc;
}
void ImProcFunctions::ShrinkAll_info(float ** WavCoeffs_a, float ** WavCoeffs_b,
int W_ab, int H_ab, float **noisevarlum, float **noisevarchrom, float **noisevarhue, float & chaut, int &Nb, float & redaut, float & blueaut,
float & maxredaut, float & maxblueaut, float & minredaut, float & minblueaut, int schoice, int lvl, float & chromina, float & sigma, float & lumema, float & sigma_L, float & redyel, float & skinc, float & nsknc,
float & maxchred, float & maxchblue, float & minchred, float & minchblue, int &nb, float & chau, float & chred, float & chblue, bool denoiseMethodRgb)
{
//simple wavelet shrinkage
if (lvl == 1) { //only one time
float chro = 0.f;
float dev = 0.f;
float devL = 0.f;
int nc = 0;
int nL = 0;
int nry = 0;
float lume = 0.f;
float red_yel = 0.f;
float skin_c = 0.f;
int nsk = 0;
for (int i = 0; i < H_ab; ++i) {
for (int j = 0; j < W_ab; ++j) {
chro += noisevarchrom[i][j];
++nc;
dev += SQR(noisevarchrom[i][j] - (chro / nc));
if (noisevarhue[i][j] > -0.8f && noisevarhue[i][j] < 2.0f && noisevarchrom[i][j] > 10000.f) {//saturated red yellow
red_yel += noisevarchrom[i][j];
++nry;
}
if (noisevarhue[i][j] > 0.f && noisevarhue[i][j] < 1.6f && noisevarchrom[i][j] < 10000.f) {//skin
skin_c += noisevarchrom[i][j];
++nsk;
}
lume += noisevarlum[i][j];
++nL;
devL += SQR(noisevarlum[i][j] - (lume / nL));
}
}
if (nc > 0) {
chromina = chro / nc;
sigma = sqrt(dev / nc);
nsknc = static_cast<float>(nsk) / static_cast<float>(nc);
} else {
nsknc = static_cast<float>(nsk);
}
if (nL > 0) {
lumema = lume / nL;
sigma_L = sqrt(devL / nL);
}
if (nry > 0) {
redyel = red_yel / nry;
}
if (nsk > 0) {
skinc = skin_c / nsk;
}
}
const float reduc = (schoice == 2) ? static_cast<float>(settings->nrhigh) : 1.f;
for (int dir = 1; dir < 4; ++dir) {
float mada, madb;
if (!denoiseMethodRgb) {
mada = SQR(Mad(WavCoeffs_a[dir], W_ab * H_ab));
} else {
mada = SQR(MadRgb(WavCoeffs_a[dir], W_ab * H_ab));
}
chred += mada;
if (mada > maxchred) {
maxchred = mada;
}
if (mada < minchred) {
minchred = mada;
}
maxredaut = sqrt(reduc * maxchred);
minredaut = sqrt(reduc * minchred);
if (!denoiseMethodRgb) {
madb = SQR(Mad(WavCoeffs_b[dir], W_ab * H_ab));
} else {
madb = SQR(MadRgb(WavCoeffs_b[dir], W_ab * H_ab));
}
chblue += madb;
if (madb > maxchblue) {
maxchblue = madb;
}
if (madb < minchblue) {
minchblue = madb;
}
maxblueaut = sqrt(reduc * maxchblue);
minblueaut = sqrt(reduc * minchblue);
chau += (mada + madb);
++nb;
//here evaluation of automatic
chaut = sqrt(reduc * chau / (nb + nb));
redaut = sqrt(reduc * chred / nb);
blueaut = sqrt(reduc * chblue / nb);
Nb = nb;
}
}
void ImProcFunctions::WaveletDenoiseAll_info(int levwav, const wavelet_decomposition & WaveletCoeffs_a,
const wavelet_decomposition & WaveletCoeffs_b, float **noisevarlum, float **noisevarchrom, float **noisevarhue, float & chaut, int &Nb, float & redaut, float & blueaut, float & maxredaut, float & maxblueaut, float & minredaut, float & minblueaut, int schoice,
float & chromina, float & sigma, float & lumema, float & sigma_L, float & redyel, float & skinc, float & nsknc, float & maxchred, float & maxchblue, float & minchred, float & minchblue, int &nb, float & chau, float & chred, float & chblue, bool denoiseMethodRgb)
{
int maxlvl = levwav;
for (int lvl = 0; lvl < maxlvl; ++lvl) {
int Wlvl_ab = WaveletCoeffs_a.level_W(lvl);
int Hlvl_ab = WaveletCoeffs_a.level_H(lvl);
float ** WavCoeffs_a = WaveletCoeffs_a.level_coeffs(lvl);
float ** WavCoeffs_b = WaveletCoeffs_b.level_coeffs(lvl);
ShrinkAll_info(WavCoeffs_a, WavCoeffs_b, Wlvl_ab, Hlvl_ab,
noisevarlum, noisevarchrom, noisevarhue, chaut, Nb, redaut, blueaut, maxredaut, maxblueaut, minredaut, minblueaut,
schoice, lvl, chromina, sigma, lumema, sigma_L, redyel, skinc, nsknc, maxchred, maxchblue, minchred, minchblue, nb, chau, chred, chblue, denoiseMethodRgb);
}
}
void ImProcFunctions::RGB_denoise_infoGamCurve(const procparams::DirPyrDenoiseParams & dnparams, bool isRAW, LUTf &gamcurve, float &gam, float &gamthresh, float &gamslope)
{
gam = dnparams.gamma;
gamthresh = 0.001f;
if (!isRAW) {//reduce gamma under 1 for Lab mode ==> TIF and JPG
if (gam < 1.9f) {
gam = 1.f - (1.9f - gam) / 3.f; //minimum gamma 0.7
} else if (gam >= 1.9f && gam <= 3.f) {
gam = (1.4f / 1.1f) * gam - 1.41818f;
}
}
bool denoiseMethodRgb = (dnparams.dmethod == "RGB");
if (denoiseMethodRgb) {
gamslope = exp(log(static_cast<double>(gamthresh)) / gam) / gamthresh;
Color::gammaf2lut(gamcurve, gam, gamthresh, gamslope, 65535.f, 32768.f);
} else {
Color::gammanf2lut(gamcurve, gam, 65535.f, 32768.f);
}
}
void ImProcFunctions::calcautodn_info(float & chaut, float & delta, int Nb, int levaut, float maxmax, float lumema, float chromina, int mode, int lissage, float redyel, float skinc, float nsknc)
{
float reducdelta = 1.f;
if (params->dirpyrDenoise.smethod == "shalbi") {
reducdelta = static_cast<float>(settings->nrhigh);
}
chaut = (chaut * Nb - maxmax) / (Nb - 1); //suppress maximum for chaut calcul
if ((redyel > 5000.f || skinc > 1000.f) && nsknc < 0.4f && chromina > 3000.f) {
chaut *= 0.45f; //reduct action in red zone, except skin for high / med chroma
} else if ((redyel > 12000.f || skinc > 1200.f) && nsknc < 0.3f && chromina > 3000.f) {
chaut *= 0.3f;
}
if (mode == 0 || mode == 2) { //Preview or Auto multizone
if (chromina > 10000.f) {
chaut *= 0.7f; //decrease action for high chroma (visible noise)
} else if (chromina > 6000.f) {
chaut *= 0.9f;
} else if (chromina < 3000.f) {
chaut *= 1.2f; //increase action in low chroma==> 1.2 /==>2.0 ==> curve CC
} else if (chromina < 2000.f) {
chaut *= 1.5f; //increase action in low chroma==> 1.5 / ==>2.7
}
if (lumema < 2500.f) {
chaut *= 1.3f; //increase action for low light
} else if (lumema < 5000.f) {
chaut *= 1.2f;
} else if (lumema > 20000.f) {
chaut *= 0.9f; //decrease for high light
}
} else if (mode == 1) {//auto ==> less coefficient because interaction
if (chromina > 10000.f) {
chaut *= 0.8f; //decrease action for high chroma (visible noise)
} else if (chromina > 6000.f) {
chaut *= 0.9f;
} else if (chromina < 3000.f) {
chaut *= 1.5f; //increase action in low chroma
} else if (chromina < 2000.f) {
chaut *= 2.2f; //increase action in low chroma
}
if (lumema < 2500.f) {
chaut *= 1.2f; //increase action for low light
} else if (lumema < 5000.f) {
chaut *= 1.1f;
} else if (lumema > 20000.f) {
chaut *= 0.9f; //decrease for high light
}
}
if (levaut == 0) { //Low denoise
if (chaut > 300.f) {
chaut = 0.714286f * chaut + 85.71428f;
}
}
delta = maxmax - chaut;
delta *= reducdelta;
if (lissage == 1 || lissage == 2) {
if (chaut < 200.f && delta < 200.f) {
delta *= 0.95f;
} else if (chaut < 200.f && delta < 400.f) {
delta *= 0.5f;
} else if (chaut < 200.f && delta >= 400.f) {
delta = 200.f;
} else if (chaut < 400.f && delta < 400.f) {
delta *= 0.4f;
} else if (chaut < 400.f && delta >= 400.f) {
delta = 120.f;
} else if (chaut < 550.f) {
delta *= 0.15f;
} else if (chaut < 650.f) {
delta *= 0.1f;
} else { /*if (chaut >= 650.f)*/
delta *= 0.07f;
}
if (mode == 0 || mode == 2) { //Preview or Auto multizone
if (chromina < 6000.f) {
delta *= 1.4f; //increase maxi
}
if (lumema < 5000.f) {
delta *= 1.4f;
}
} else if (mode == 1) { //Auto
if (chromina < 6000.f) {
delta *= 1.2f; //increase maxi
}
if (lumema < 5000.f) {
delta *= 1.2f;
}
}
}
if (lissage == 0) {
if (chaut < 200.f && delta < 200.f) {
delta *= 0.95f;
} else if (chaut < 200.f && delta < 400.f) {
delta *= 0.7f;
} else if (chaut < 200.f && delta >= 400.f) {
delta = 280.f;
} else if (chaut < 400.f && delta < 400.f) {
delta *= 0.6f;
} else if (chaut < 400.f && delta >= 400.f) {
delta = 200.f;
} else if (chaut < 550.f) {
delta *= 0.3f;
} else if (chaut < 650.f) {
delta *= 0.2f;
} else { /*if (chaut >= 650.f)*/
delta *= 0.15f;
}
if (mode == 0 || mode == 2) { //Preview or Auto multizone
if (chromina < 6000.f) {
delta *= 1.4f; //increase maxi
}
if (lumema < 5000.f) {
delta *= 1.4f;
}
} else if (mode == 1) { //Auto
if (chromina < 6000.f) {
delta *= 1.2f; //increase maxi
}
if (lumema < 5000.f) {
delta *= 1.2f;
}
}
}
}
void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat * provicalc, const bool isRAW, const LUTf &gamcurve, float gam, float gamthresh, float gamslope, const procparams::DirPyrDenoiseParams & dnparams, const double expcomp, float &chaut, int &Nb, float &redaut, float &blueaut, float &maxredaut, float &maxblueaut, float &minredaut, float &minblueaut, float &chromina, float &sigma, float &lumema, float &sigma_L, float &redyel, float &skinc, float &nsknc, bool multiThread)
{
if ((settings->leveldnautsimpl == 1 && dnparams.Cmethod == "MAN") || (settings->leveldnautsimpl == 0 && dnparams.C2method == "MANU")) {
//nothing to do
return;
}
int hei, wid;
float** lumcalc;
float** acalc;
float** bcalc;
hei = provicalc->getHeight();
wid = provicalc->getWidth();
TMatrix wprofi = ICCStore::getInstance()->workingSpaceMatrix(params->icm.workingProfile);
const float wpi[3][3] = {
{static_cast<float>(wprofi[0][0]), static_cast<float>(wprofi[0][1]), static_cast<float>(wprofi[0][2])},
{static_cast<float>(wprofi[1][0]), static_cast<float>(wprofi[1][1]), static_cast<float>(wprofi[1][2])},
{static_cast<float>(wprofi[2][0]), static_cast<float>(wprofi[2][1]), static_cast<float>(wprofi[2][2])}
};
lumcalc = new float*[hei];
for (int i = 0; i < hei; ++i) {
lumcalc[i] = new float[wid];
}
acalc = new float*[hei];
for (int i = 0; i < hei; ++i) {
acalc[i] = new float[wid];
}
bcalc = new float*[hei];
for (int i = 0; i < hei; ++i) {
bcalc[i] = new float[wid];
}
#ifdef _OPENMP
#pragma omp parallel for if (multiThread)
#endif
for (int ii = 0; ii < hei; ++ii) {
for (int jj = 0; jj < wid; ++jj) {
float LLum, AAum, BBum;
float RL = provicalc->r(ii, jj);
float GL = provicalc->g(ii, jj);
float BL = provicalc->b(ii, jj);
// determine luminance for noisecurve
float XL, YL, ZL;
Color::rgbxyz(RL, GL, BL, XL, YL, ZL, wpi);
Color::XYZ2Lab(XL, YL, ZL, LLum, AAum, BBum);
lumcalc[ii][jj] = LLum;
acalc[ii][jj] = AAum;
bcalc[ii][jj] = BBum;
}
}
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
const int imheight = src->getHeight(), imwidth = src->getWidth();
bool denoiseMethodRgb = (dnparams.dmethod == "RGB");
const float gain = pow(2.0f, float(expcomp));
int tilesize = 0;
int overlap = 0;
if (settings->leveldnti == 0) {
tilesize = 1024;
overlap = 128;
}
if (settings->leveldnti == 1) {
tilesize = 768;
overlap = 96;
}
int numtiles_W, numtiles_H, tilewidth, tileheight, tileWskip, tileHskip;
//always no Tiles
int kall = 0;
Tile_calc(tilesize, overlap, kall, imwidth, imheight, numtiles_W, numtiles_H, tilewidth, tileheight, tileWskip, tileHskip);
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
TMatrix wprof = ICCStore::getInstance()->workingSpaceMatrix(params->icm.workingProfile);
const float wp[3][3] = {
{static_cast<float>(wprof[0][0]), static_cast<float>(wprof[0][1]), static_cast<float>(wprof[0][2])},
{static_cast<float>(wprof[1][0]), static_cast<float>(wprof[1][1]), static_cast<float>(wprof[1][2])},
{static_cast<float>(wprof[2][0]), static_cast<float>(wprof[2][1]), static_cast<float>(wprof[2][2])}
};
float chau = 0.f;
float chred = 0.f;
float chblue = 0.f;
float maxchred = 0.f;
float maxchblue = 0.f;
float minchred = 100000000.f;
float minchblue = 100000000.f;
int nb = 0;
int comptlevel = 0;
for (int tiletop = 0; tiletop < imheight; tiletop += tileHskip) {
for (int tileleft = 0; tileleft < imwidth; tileleft += tileWskip) {
int tileright = MIN(imwidth, tileleft + tilewidth);
int tilebottom = MIN(imheight, tiletop + tileheight);
int width = tileright - tileleft;
int height = tilebottom - tiletop;
LabImage * labdn = new LabImage(width, height);
float** noisevarlum = new float*[(height + 1) / 2];
for (int i = 0; i < (height + 1) / 2; ++i) {
noisevarlum[i] = new float[(width + 1) / 2];
}
float** noisevarchrom = new float*[(height + 1) / 2];
for (int i = 0; i < (height + 1) / 2; ++i) {
noisevarchrom[i] = new float[(width + 1) / 2];
}
float** noisevarhue = new float*[(height + 1) / 2];
for (int i = 0; i < (height + 1) / 2; ++i) {
noisevarhue[i] = new float[(width + 1) / 2];
}
float realred, realblue;
float interm_med = static_cast<float>(dnparams.chroma) / 10.0;
float intermred, intermblue;
if (dnparams.redchro > 0.) {
intermred = (dnparams.redchro / 10.);
} else {
intermred = static_cast<float>(dnparams.redchro) / 7.0; //increase slower than linear for more sensit
}
if (dnparams.bluechro > 0.) {
intermblue = (dnparams.bluechro / 10.);
} else {
intermblue = static_cast<float>(dnparams.bluechro) / 7.0; //increase slower than linear for more sensit
}
realred = interm_med + intermred;
if (realred < 0.f) {
realred = 0.001f;
}
realblue = interm_med + intermblue;
if (realblue < 0.f) {
realblue = 0.001f;
}
//fill tile from image; convert RGB to "luma/chroma"
if (isRAW) {//image is raw; use channel differences for chroma channels
#ifdef _OPENMP
#pragma omp parallel for if (multiThread)
#endif
for (int i = tiletop; i < tilebottom; i += 2) {
int i1 = i - tiletop;
#ifdef __SSE2__
vfloat aNv, bNv;
vfloat c100v = F2V(100.f);
int j;
for (j = tileleft; j < tileright - 7; j += 8) {
int j1 = j - tileleft;
aNv = LVFU(acalc[i >> 1][j >> 1]);
bNv = LVFU(bcalc[i >> 1][j >> 1]);
STVFU(noisevarhue[i1 >> 1][j1 >> 1], xatan2f(bNv, aNv));
STVFU(noisevarchrom[i1 >> 1][j1 >> 1], vmaxf(vsqrtf(SQRV(aNv) + SQRV(bNv)),c100v));
}
for (; j < tileright; j += 2) {
int j1 = j - tileleft;
float aN = acalc[i >> 1][j >> 1];
float bN = bcalc[i >> 1][j >> 1];
float cN = sqrtf(SQR(aN) + SQR(bN));
noisevarhue[i1 >> 1][j1 >> 1] = xatan2f(bN, aN);
if (cN < 100.f) {
cN = 100.f; //avoid divided by zero
}
noisevarchrom[i1 >> 1][j1 >> 1] = cN;
}
#else
for (int j = tileleft; j < tileright; j += 2) {
int j1 = j - tileleft;
float aN = acalc[i >> 1][j >> 1];
float bN = bcalc[i >> 1][j >> 1];
float cN = sqrtf(SQR(aN) + SQR(bN));
float hN = xatan2f(bN, aN);
if (cN < 100.f) {
cN = 100.f; //avoid divided by zero
}
noisevarchrom[i1 >> 1][j1 >> 1] = cN;
noisevarhue[i1 >> 1][j1 >> 1] = hN;
}
#endif
}
#ifdef _OPENMP
#pragma omp parallel for if (multiThread)
#endif
for (int i = tiletop; i < tilebottom; i += 2) {
int i1 = i - tiletop;
for (int j = tileleft; j < tileright; j += 2) {
int j1 = j - tileleft;
float Llum = lumcalc[i >> 1][j >> 1];
Llum = Llum < 2.f ? 2.f : Llum; //avoid divided by zero ?
Llum = Llum > 32768.f ? 32768.f : Llum; // not strictly necessary
noisevarlum[i1 >> 1][j1 >> 1] = Llum;
}
}
if (!denoiseMethodRgb) { //lab mode, modification Jacques feb 2013 and july 2014
#ifdef _OPENMP
#pragma omp parallel for if (multiThread)
#endif
for (int i = tiletop; i < tilebottom; ++i) {
int i1 = i - tiletop;
for (int j = tileleft; j < tileright; ++j) {
int j1 = j - tileleft;
float R_ = gain * src->r(i, j);
float G_ = gain * src->g(i, j);
float B_ = gain * src->b(i, j);
R_ = Color::denoiseIGammaTab[R_];
G_ = Color::denoiseIGammaTab[G_];
B_ = Color::denoiseIGammaTab[B_];
//apply gamma noise standard (slider)
R_ = R_ < 65535.f ? gamcurve[R_] : (Color::gammanf(R_ / 65535.f, gam) * 32768.f);
G_ = G_ < 65535.f ? gamcurve[G_] : (Color::gammanf(G_ / 65535.f, gam) * 32768.f);
B_ = B_ < 65535.f ? gamcurve[B_] : (Color::gammanf(B_ / 65535.f, gam) * 32768.f);
//true conversion xyz=>Lab
float X, Y, Z;
Color::rgbxyz(R_, G_, B_, X, Y, Z, wp);
//convert to Lab
float L, a, b;
Color::XYZ2Lab(X, Y, Z, L, a, b);
labdn->a[i1][j1] = a;
labdn->b[i1][j1] = b;
}
}
} else { //RGB mode
for (int i = tiletop/*, i1=0*/; i < tilebottom; ++i/*, ++i1*/) {
int i1 = i - tiletop;
for (int j = tileleft/*, j1=0*/; j < tileright; ++j/*, ++j1*/) {
int j1 = j - tileleft;
float X = gain * src->r(i, j);
float Y = gain * src->g(i, j);
float Z = gain * src->b(i, j);
X = X < 65535.f ? gamcurve[X] : (Color::gammaf(X / 65535.f, gam, gamthresh, gamslope) * 32768.f);
Y = Y < 65535.f ? gamcurve[Y] : (Color::gammaf(Y / 65535.f, gam, gamthresh, gamslope) * 32768.f);
Z = Z < 65535.f ? gamcurve[Z] : (Color::gammaf(Z / 65535.f, gam, gamthresh, gamslope) * 32768.f);
labdn->a[i1][j1] = (X - Y);
labdn->b[i1][j1] = (Y - Z);
}
}
}
} else {//image is not raw; use Lab parametrization
for (int i = tiletop/*, i1=0*/; i < tilebottom; ++i/*, ++i1*/) {
int i1 = i - tiletop;
for (int j = tileleft/*, j1=0*/; j < tileright; ++j/*, ++j1*/) {
int j1 = j - tileleft;
float L, a, b;
float rLum = src->r(i, j) ; //for luminance denoise curve
float gLum = src->g(i, j) ;
float bLum = src->b(i, j) ;
//use gamma sRGB, not good if TIF (JPG) Output profil not with gamma sRGB (eg : gamma =1.0, or 1.8...)
//very difficult to solve !
// solution ==> save TIF with gamma sRGB and re open
float rtmp = Color::igammatab_srgb[ src->r(i, j) ];
float gtmp = Color::igammatab_srgb[ src->g(i, j) ];
float btmp = Color::igammatab_srgb[ src->b(i, j) ];
//modification Jacques feb 2013
// gamma slider different from raw
rtmp = rtmp < 65535.f ? gamcurve[rtmp] : (Color::gammanf(rtmp / 65535.f, gam) * 32768.f);
gtmp = gtmp < 65535.f ? gamcurve[gtmp] : (Color::gammanf(gtmp / 65535.f, gam) * 32768.f);
btmp = btmp < 65535.f ? gamcurve[btmp] : (Color::gammanf(btmp / 65535.f, gam) * 32768.f);
float X, Y, Z;
Color::rgbxyz(rtmp, gtmp, btmp, X, Y, Z, wp);
//convert Lab
Color::XYZ2Lab(X, Y, Z, L, a, b);
if (((i1 | j1) & 1) == 0) {
float Llum, alum, blum;
float XL, YL, ZL;
Color::rgbxyz(rLum, gLum, bLum, XL, YL, ZL, wp);
Color::XYZ2Lab(XL, YL, ZL, Llum, alum, blum);
float kN = Llum;
if (kN < 2.f) {
kN = 2.f;
}
if (kN > 32768.f) {
kN = 32768.f;
}
noisevarlum[i1 >> 1][j1 >> 1] = kN;
float aN = alum;
float bN = blum;
float hN = xatan2f(bN, aN);
float cN = sqrt(SQR(aN) + SQR(bN));
if (cN < 100.f) {
cN = 100.f; //avoid divided by zero
}
noisevarchrom[i1 >> 1][j1 >> 1] = cN;
noisevarhue[i1 >> 1][j1 >> 1] = hN;
}
labdn->a[i1][j1] = a;
labdn->b[i1][j1] = b;
}
}
}
int datalen = labdn->W * labdn->H;
//now perform basic wavelet denoise
//last two arguments of wavelet decomposition are max number of wavelet decomposition levels;
//and whether to subsample the image after wavelet filtering. Subsampling is coded as
//binary 1 or 0 for each level, eg subsampling = 0 means no subsampling, 1 means subsample
//the first level only, 7 means subsample the first three levels, etc.
wavelet_decomposition* adecomp;
wavelet_decomposition* bdecomp;
int schoice = 0;//shrink method
if (dnparams.smethod == "shalbi") {
schoice = 2;
}
const int levwav = 5;
#ifdef _OPENMP
#pragma omp parallel sections if (multiThread)
#endif
{
#ifdef _OPENMP
#pragma omp section
#endif
{
adecomp = new wavelet_decomposition(labdn->data + datalen, labdn->W, labdn->H, levwav, 1);
}
#ifdef _OPENMP
#pragma omp section
#endif
{
bdecomp = new wavelet_decomposition(labdn->data + 2 * datalen, labdn->W, labdn->H, levwav, 1);
}
}
if (comptlevel == 0) {
WaveletDenoiseAll_info(
levwav,
*adecomp,
*bdecomp,
noisevarlum,
noisevarchrom,
noisevarhue,
chaut,
Nb,
redaut,
blueaut,
maxredaut,
maxblueaut,
minredaut,
minblueaut,
schoice,
chromina,
sigma,
lumema,
sigma_L,
redyel,
skinc,
nsknc,
maxchred,
maxchblue,
minchred,
minchblue,
nb,
chau,
chred,
chblue,
denoiseMethodRgb
); // Enhance mode
}
comptlevel += 1;
delete adecomp;
delete bdecomp;
delete labdn;
for (int i = 0; i < (height + 1) / 2; ++i) {
delete[] noisevarlum[i];
}
delete[] noisevarlum;
for (int i = 0; i < (height + 1) / 2; ++i) {
delete[] noisevarchrom[i];
}
delete[] noisevarchrom;
for (int i = 0; i < (height + 1) / 2; ++i) {
delete[] noisevarhue[i];
}
delete[] noisevarhue;
}//end of tile row
}//end of tile loop
for (int i = 0; i < hei; ++i) {
delete[] lumcalc[i];
}
delete[] lumcalc;
for (int i = 0; i < hei; ++i) {
delete[] acalc[i];
}
delete[] acalc;
for (int i = 0; i < hei; ++i) {
delete[] bcalc[i];
}
delete[] bcalc;
#undef TS
#undef fTS
#undef offset
#undef epsilon
} // End of main RGB_denoise
}
Reference in New Issue View Git Blame Copy Permalink