liz/rawTherapee
liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
rawTherapee/rtengine/rawflatfield.cc
Ingo Weyrich ccf4b4c843 flatfield: code review
2019-10-23 23:14:08 +02:00

561 lines
24 KiB
C++
Raw Blame History

/*
* This file is part of RawTherapee.
*
* Copyright (c) 2004-2017 Gabor Horvath <hgabor@rawtherapee.com>
*
* RawTherapee 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.
*
* RawTherapee 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 RawTherapee. If not, see <https://www.gnu.org/licenses/>.
*/
#include <cmath>
#include <new>
#include "rtengine.h"
#include "rawimagesource.h"
#include "rawimage.h"
#include "procparams.h"
//#define BENCHMARK
//#include "StopWatch.h"
#include "opthelper.h"
namespace {
void cfaboxblur(float** riFlatFile, float* cfablur, int boxH, int boxW, int H, int W)
{
if (boxW < 0 || boxH < 0 || (boxW == 0 && boxH == 0)) { // nothing to blur or negative values
memcpy(cfablur, riFlatFile[0], W * H * sizeof(float));
return;
}
float *tmpBuffer = nullptr;
float *cfatmp = nullptr;
float *srcVertical = nullptr;
if (boxH > 0 && boxW > 0) {
// we need a temporary buffer if we have to blur both directions
tmpBuffer = (float (*)) calloc (H * W, sizeof * tmpBuffer);
}
if (boxH == 0) {
// if boxH == 0 we can skip the vertical blur and process the horizontal blur from riFlatFile to cfablur without using a temporary buffer
cfatmp = cfablur;
} else {
cfatmp = tmpBuffer;
}
if (boxW == 0) {
// if boxW == 0 we can skip the horizontal blur and process the vertical blur from riFlatFile to cfablur without using a temporary buffer
srcVertical = riFlatFile[0];
} else {
srcVertical = cfatmp;
}
#ifdef _OPENMP
#pragma omp parallel
#endif
{
if (boxW > 0) {
//box blur cfa image; box size = BS
//horizontal blur
#ifdef _OPENMP
#pragma omp for
#endif
for (int row = 0; row < H; ++row) {
int len = boxW / 2 + 1;
cfatmp[row * W + 0] = riFlatFile[row][0] / len;
cfatmp[row * W + 1] = riFlatFile[row][1] / len;
for (int j = 2; j <= boxW; j += 2) {
cfatmp[row * W + 0] += riFlatFile[row][j] / len;
cfatmp[row * W + 1] += riFlatFile[row][j + 1] / len;
}
for (int col = 2; col <= boxW; col += 2) {
cfatmp[row * W + col] = (cfatmp[row * W + col - 2] * len + riFlatFile[row][boxW + col]) / (len + 1);
cfatmp[row * W + col + 1] = (cfatmp[row * W + col - 1] * len + riFlatFile[row][boxW + col + 1]) / (len + 1);
len ++;
}
for (int col = boxW + 2; col < W - boxW; col++) {
cfatmp[row * W + col] = cfatmp[row * W + col - 2] + (riFlatFile[row][boxW + col] - cfatmp[row * W + col - boxW - 2]) / len;
}
for (int col = W - boxW; col < W; col += 2) {
cfatmp[row * W + col] = (cfatmp[row * W + col - 2] * len - cfatmp[row * W + col - boxW - 2]) / (len - 1);
if (col + 1 < W) {
cfatmp[row * W + col + 1] = (cfatmp[row * W + col - 1] * len - cfatmp[row * W + col - boxW - 1]) / (len - 1);
}
len --;
}
}
}
if (boxH > 0) {
//vertical blur
#ifdef __SSE2__
const vfloat leninitv = F2V(boxH / 2 + 1);
const vfloat onev = F2V(1.f);
#ifdef _OPENMP
#pragma omp for nowait
#endif
for (int col = 0; col < W - 7; col += 8) {
vfloat lenv = leninitv;
vfloat temp1v = LVFU(srcVertical[0 * W + col]) / lenv;
vfloat temp2v = LVFU(srcVertical[1 * W + col]) / lenv;
vfloat temp3v = LVFU(srcVertical[0 * W + col + 4]) / lenv;
vfloat temp4v = LVFU(srcVertical[1 * W + col + 4]) / lenv;
for (int i = 2; i < boxH + 2; i += 2) {
temp1v += LVFU(srcVertical[i * W + col]) / lenv;
temp2v += LVFU(srcVertical[(i + 1) * W + col]) / lenv;
temp3v += LVFU(srcVertical[i * W + col + 4]) / lenv;
temp4v += LVFU(srcVertical[(i + 1) * W + col + 4]) / lenv;
}
STVFU(cfablur[0 * W + col], temp1v);
STVFU(cfablur[1 * W + col], temp2v);
STVFU(cfablur[0 * W + col + 4], temp3v);
STVFU(cfablur[1 * W + col + 4], temp4v);
int row;
for (row = 2; row < boxH + 2; row += 2) {
const vfloat lenp1v = lenv + onev;
temp1v = (temp1v * lenv + LVFU(srcVertical[(row + boxH) * W + col])) / lenp1v;
temp2v = (temp2v * lenv + LVFU(srcVertical[(row + boxH + 1) * W + col])) / lenp1v;
temp3v = (temp3v * lenv + LVFU(srcVertical[(row + boxH) * W + col + 4])) / lenp1v;
temp4v = (temp4v * lenv + LVFU(srcVertical[(row + boxH + 1) * W + col + 4])) / lenp1v;
STVFU(cfablur[row * W + col], temp1v);
STVFU(cfablur[(row + 1)*W + col], temp2v);
STVFU(cfablur[row * W + col + 4], temp3v);
STVFU(cfablur[(row + 1)*W + col + 4], temp4v);
lenv = lenp1v;
}
for (; row < H - boxH - 1; row += 2) {
temp1v = temp1v + (LVFU(srcVertical[(row + boxH) * W + col]) - LVFU(srcVertical[(row - boxH - 2) * W + col])) / lenv;
temp2v = temp2v + (LVFU(srcVertical[(row + 1 + boxH) * W + col]) - LVFU(srcVertical[(row + 1 - boxH - 2) * W + col])) / lenv;
temp3v = temp3v + (LVFU(srcVertical[(row + boxH) * W + col + 4]) - LVFU(srcVertical[(row - boxH - 2) * W + col + 4])) / lenv;
temp4v = temp4v + (LVFU(srcVertical[(row + 1 + boxH) * W + col + 4]) - LVFU(srcVertical[(row + 1 - boxH - 2) * W + col + 4])) / lenv;
STVFU(cfablur[row * W + col], temp1v);
STVFU(cfablur[(row + 1)*W + col], temp2v);
STVFU(cfablur[row * W + col + 4], temp3v);
STVFU(cfablur[(row + 1)*W + col + 4], temp4v);
}
if (row < H - boxH) {
temp1v = temp1v + (LVFU(srcVertical[(row + boxH) * W + col]) - LVFU(srcVertical[(row - boxH - 2) * W + col])) / lenv;
temp3v = temp3v + (LVFU(srcVertical[(row + boxH) * W + col + 4]) - LVFU(srcVertical[(row - boxH - 2) * W + col + 4])) / lenv;
STVFU(cfablur[row * W + col], temp1v);
STVFU(cfablur[row * W + col + 4], temp3v);
vfloat swapv = temp1v;
temp1v = temp2v;
temp2v = swapv;
swapv = temp3v;
temp3v = temp4v;
temp4v = swapv;
++row;
}
for (; row < H - 1; row += 2) {
const vfloat lenm1v = lenv - onev;
temp1v = (temp1v * lenv - LVFU(srcVertical[(row - boxH - 2) * W + col])) / lenm1v;
temp2v = (temp2v * lenv - LVFU(srcVertical[(row - boxH - 1) * W + col])) / lenm1v;
temp3v = (temp3v * lenv - LVFU(srcVertical[(row - boxH - 2) * W + col + 4])) / lenm1v;
temp4v = (temp4v * lenv - LVFU(srcVertical[(row - boxH - 1) * W + col + 4])) / lenm1v;
STVFU(cfablur[row * W + col], temp1v);
STVFU(cfablur[(row + 1)*W + col], temp2v);
STVFU(cfablur[row * W + col + 4], temp3v);
STVFU(cfablur[(row + 1)*W + col + 4], temp4v);
lenv = lenm1v;
}
if (row < H) {
vfloat lenm1v = lenv - onev;
temp1v = (temp1v * lenv - LVFU(srcVertical[(row - boxH - 2) * W + col])) / lenm1v;
temp3v = (temp3v * lenv - LVFU(srcVertical[(row - boxH - 2) * W + col + 4])) / lenm1v;
STVFU(cfablur[(row)*W + col], temp1v);
STVFU(cfablur[(row)*W + col + 4], temp3v);
}
}
#ifdef _OPENMP
#pragma omp single
#endif
for (int col = W - (W % 8); col < W; ++col) {
int len = boxH / 2 + 1;
cfablur[0 * W + col] = srcVertical[0 * W + col] / len;
cfablur[1 * W + col] = srcVertical[1 * W + col] / len;
for (int i = 2; i < boxH + 2; i += 2) {
cfablur[0 * W + col] += srcVertical[i * W + col] / len;
cfablur[1 * W + col] += srcVertical[(i + 1) * W + col] / len;
}
for (int row = 2; row < boxH + 2; row += 2) {
cfablur[row * W + col] = (cfablur[(row - 2) * W + col] * len + srcVertical[(row + boxH) * W + col]) / (len + 1);
cfablur[(row + 1)*W + col] = (cfablur[(row - 1) * W + col] * len + srcVertical[(row + boxH + 1) * W + col]) / (len + 1);
len ++;
}
for (int row = boxH + 2; row < H - boxH; ++row) {
cfablur[row * W + col] = cfablur[(row - 2) * W + col] + (srcVertical[(row + boxH) * W + col] - srcVertical[(row - boxH - 2) * W + col]) / len;
}
for (int row = H - boxH; row < H; row += 2) {
cfablur[row * W + col] = (cfablur[(row - 2) * W + col] * len - srcVertical[(row - boxH - 2) * W + col]) / (len - 1);
if (row + 1 < H) {
cfablur[(row + 1)*W + col] = (cfablur[(row - 1) * W + col] * len - srcVertical[(row - boxH - 1) * W + col]) / (len - 1);
}
len --;
}
}
#else
#ifdef _OPENMP
#pragma omp for
#endif
for (int col = 0; col < W; ++col) {
int len = boxH / 2 + 1;
cfablur[0 * W + col] = srcVertical[0 * W + col] / len;
cfablur[1 * W + col] = srcVertical[1 * W + col] / len;
for (int i = 2; i < boxH + 2; i += 2) {
cfablur[0 * W + col] += srcVertical[i * W + col] / len;
cfablur[1 * W + col] += srcVertical[(i + 1) * W + col] / len;
}
for (int row = 2; row < boxH + 2; row += 2) {
cfablur[row * W + col] = (cfablur[(row - 2) * W + col] * len + srcVertical[(row + boxH) * W + col]) / (len + 1);
cfablur[(row + 1)*W + col] = (cfablur[(row - 1) * W + col] * len + srcVertical[(row + boxH + 1) * W + col]) / (len + 1);
len ++;
}
for (int row = boxH + 2; row < H - boxH; ++row) {
cfablur[row * W + col] = cfablur[(row - 2) * W + col] + (srcVertical[(row + boxH) * W + col] - srcVertical[(row - boxH - 2) * W + col]) / len;
}
for (int row = H - boxH; row < H; row += 2) {
cfablur[row * W + col] = (cfablur[(row - 2) * W + col] * len - srcVertical[(row - boxH - 2) * W + col]) / (len - 1);
if (row + 1 < H) {
cfablur[(row + 1)*W + col] = (cfablur[(row - 1) * W + col] * len - srcVertical[(row - boxH - 1) * W + col]) / (len - 1);
}
len --;
}
}
#endif
}
}
if (tmpBuffer) {
free (tmpBuffer);
}
}
}
namespace rtengine
{
void RawImageSource::processFlatField(const RAWParams &raw, RawImage *riFlatFile, unsigned short black[4])
{
// BENCHFUN
float *cfablur = new float[H * W];
int BS = raw.ff_BlurRadius;
BS += BS & 1;
if (raw.ff_BlurType == RAWParams::getFlatFieldBlurTypeString(RAWParams::FlatFieldBlurType::V)) {
cfaboxblur(riFlatFile->data, cfablur, 2 * BS, 0, H, W);
} else if (raw.ff_BlurType == RAWParams::getFlatFieldBlurTypeString(RAWParams::FlatFieldBlurType::H)) {
cfaboxblur(riFlatFile->data, cfablur, 0, 2 * BS, H, W);
} else if (raw.ff_BlurType == RAWParams::getFlatFieldBlurTypeString(RAWParams::FlatFieldBlurType::VH)) {
//slightly more complicated blur if trying to correct both vertical and horizontal anomalies
cfaboxblur(riFlatFile->data, cfablur, BS, BS, H, W); //first do area blur to correct vignette
} else { //(raw.ff_BlurType == RAWParams::getFlatFieldBlurTypeString(RAWParams::area_ff))
cfaboxblur(riFlatFile->data, cfablur, BS, BS, H, W);
}
if (ri->getSensorType() == ST_BAYER || ri->get_colors() == 1) {
float refcolor[2][2];
// find center values by channel
for (int m = 0; m < 2; ++m)
for (int n = 0; n < 2; ++n) {
const int row = 2 * (H >> 2) + m;
const int col = 2 * (W >> 2) + n;
const int c = ri->get_colors() != 1 ? FC(row, col) : 0;
const int c4 = ri->get_colors() != 1 ? ((c == 1 && !(row & 1)) ? 3 : c) : 0;
refcolor[m][n] = std::max(0.0f, cfablur[row * W + col] - black[c4]);
}
float limitFactor = 1.f;
if (raw.ff_AutoClipControl) {
bool clippedBefore = false;
for (int m = 0; m < 2 && !clippedBefore; ++m) {
for (int n = 0; n < 2 && !clippedBefore; ++n) {
float maxval = 0.f;
const int c = ri->get_colors() != 1 ? FC(m, n) : 0;
const int c4 = ri->get_colors() != 1 ? ((c == 1 && !(m & 1)) ? 3 : c) : 0;
const float clipVal = ri->get_white(c4);
#ifdef _OPENMP
#pragma omp parallel for reduction(max:maxval) schedule(dynamic, 16)
#endif
for (int row = 0; row < H - m; row += 2) {
for (int col = 0; col < W - n && !clippedBefore; col += 2) {
const float rawVal = rawData[row + m][col + n];
if (rawVal >= clipVal) {
clippedBefore = true;
break;
}
const float tempval = (rawVal - black[c4]) * (refcolor[m][n] / std::max(1e-5f, cfablur[(row + m) * W + col + n] - black[c4]));
maxval = std::max(maxval, tempval);
}
}
// now we have the max value for the channel
// if it clips, calculate factor to avoid clipping
if (maxval + black[c4] >= ri->get_white(c4)) {
if (!clippedBefore) {
limitFactor = std::min(limitFactor, ri->get_white(c4) / (maxval + black[c4]));
} else {
limitFactor = 1.f;
}
}
}
}
flatFieldAutoClipValue = (1.f - limitFactor) * 100.f; // this value can be used to set the clip control slider in gui
} else {
limitFactor = std::max((float)(100 - raw.ff_clipControl) / 100.f, 0.01f);
}
for (int m = 0; m < 2; ++m)
for (int n = 0; n < 2; ++n) {
refcolor[m][n] *= limitFactor;
}
unsigned int c[2][2] {};
unsigned int c4[2][2] {};
if (ri->get_colors() != 1) {
for (int i = 0; i < 2; ++i) {
for (int j = 0; j < 2; ++j) {
c[i][j] = FC(i, j);
}
}
c4[0][0] = (c[0][0] == 1) ? 3 : c[0][0];
c4[0][1] = (c[0][1] == 1) ? 3 : c[0][1];
c4[1][0] = c[1][0];
c4[1][1] = c[1][1];
}
constexpr float minValue = 1.f; // if the pixel value in the flat field is less or equal this value, no correction will be applied.
#ifdef __SSE2__
const vfloat refcolorv[2] = {_mm_set_ps(refcolor[0][1], refcolor[0][0], refcolor[0][1], refcolor[0][0]),
_mm_set_ps(refcolor[1][1], refcolor[1][0], refcolor[1][1], refcolor[1][0])
};
const vfloat blackv[2] = {_mm_set_ps(black[c4[0][1]], black[c4[0][0]], black[c4[0][1]], black[c4[0][0]]),
_mm_set_ps(black[c4[1][1]], black[c4[1][0]], black[c4[1][1]], black[c4[1][0]])
};
const vfloat onev = F2V(1.f);
const vfloat minValuev = F2V(minValue);
#endif
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for (int row = 0; row < H; ++row) {
int col = 0;
#ifdef __SSE2__
const vfloat rowBlackv = blackv[row & 1];
const vfloat rowRefcolorv = refcolorv[row & 1];
for (; col < W - 3; col += 4) {
const vfloat blurv = LVFU(cfablur[(row) * W + col]) - rowBlackv;
vfloat vignettecorrv = rowRefcolorv / blurv;
vignettecorrv = vself(vmaskf_le(blurv, minValuev), onev, vignettecorrv);
const vfloat valv = LVFU(rawData[row][col]) - rowBlackv;
STVFU(rawData[row][col], valv * vignettecorrv + rowBlackv);
}
#endif
for (; col < W; ++col) {
const float blur = cfablur[(row) * W + col] - black[c4[row & 1][col & 1]];
const float vignettecorr = blur <= minValue ? 1.f : refcolor[row & 1][col & 1] / blur;
rawData[row][col] = (rawData[row][col] - black[c4[row & 1][col & 1]]) * vignettecorr + black[c4[row & 1][col & 1]];
}
}
} else if (ri->getSensorType() == ST_FUJI_XTRANS) {
float refcolor[3] = {0.f};
int cCount[3] = {0};
// find center average values by channel
for (int m = -3; m < 3; ++m)
for (int n = -3; n < 3; ++n) {
const int row = 2 * (H >> 2) + m;
const int col = 2 * (W >> 2) + n;
const int c = riFlatFile->XTRANSFC(row, col);
refcolor[c] += std::max(0.0f, cfablur[row * W + col] - black[c]);
cCount[c] ++;
}
for (int c = 0; c < 3; ++c) {
refcolor[c] = refcolor[c] / cCount[c];
}
float limitFactor = 1.f;
if (raw.ff_AutoClipControl) {
// determine maximum calculated value to avoid clipping
bool clippedBefore = false;
const float clipVal = ri->get_white(0);
float maxval = 0.f;
// xtrans files have only one black level actually, so we can simplify the code a bit
#ifdef _OPENMP
#pragma omp parallel for reduction(max:maxval) schedule(dynamic,16)
#endif
for (int row = 0; row < H; ++row) {
for (int col = 0; col < W && !clippedBefore; ++col) {
const float rawVal = rawData[row][col];
if (rawVal >= clipVal) {
clippedBefore = true;
break;
}
const float tempval = (rawVal - black[0]) * (refcolor[ri->XTRANSFC(row, col)] / std::max(1e-5f, cfablur[(row) * W + col] - black[0]));
maxval = std::max(maxval, tempval);
}
}
// there's only one white level for xtrans
if (!clippedBefore && maxval + black[0] > ri->get_white(0)) {
limitFactor = ri->get_white(0) / (maxval + black[0]);
flatFieldAutoClipValue = (1.f - limitFactor) * 100.f; // this value can be used to set the clip control slider in gui
}
} else {
limitFactor = std::max((float)(100 - raw.ff_clipControl) / 100.f, 0.01f);
}
for (int c = 0; c < 3; ++c) {
refcolor[c] *= limitFactor;
}
constexpr float minValue = 1.f; // if the pixel value in the flat field is less or equal this value, no correction will be applied.
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; ++row) {
for (int col = 0; col < W; ++col) {
const int c = ri->XTRANSFC(row, col);
const float blur = cfablur[(row) * W + col] - black[c];
const float vignettecorr = blur <= minValue ? 1.f : refcolor[c] / blur;
rawData[row][col] = (rawData[row][col] - black[c]) * vignettecorr + black[c];
}
}
}
if (raw.ff_BlurType == RAWParams::getFlatFieldBlurTypeString(RAWParams::FlatFieldBlurType::VH)) {
float *cfablur1 = new float[H * W];
float *cfablur2 = new float[H * W];
//slightly more complicated blur if trying to correct both vertical and horizontal anomalies
cfaboxblur(riFlatFile->data, cfablur1, 0, 2 * BS, H, W); //now do horizontal blur
cfaboxblur(riFlatFile->data, cfablur2, 2 * BS, 0, H, W); //now do vertical blur
if (ri->getSensorType() == ST_BAYER || ri->get_colors() == 1) {
unsigned int c[2][2] {};
unsigned int c4[2][2] {};
if (ri->get_colors() != 1) {
for (int i = 0; i < 2; ++i) {
for (int j = 0; j < 2; ++j) {
c[i][j] = FC(i, j);
}
}
c4[0][0] = (c[0][0] == 1) ? 3 : c[0][0];
c4[0][1] = (c[0][1] == 1) ? 3 : c[0][1];
c4[1][0] = c[1][0];
c4[1][1] = c[1][1];
}
#ifdef __SSE2__
const vfloat blackv[2] = {_mm_set_ps(black[c4[0][1]], black[c4[0][0]], black[c4[0][1]], black[c4[0][0]]),
_mm_set_ps(black[c4[1][1]], black[c4[1][0]], black[c4[1][1]], black[c4[1][0]])
};
const vfloat epsv = F2V(1e-5f);
#endif
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for (int row = 0; row < H; ++row) {
int col = 0;
#ifdef __SSE2__
const vfloat rowBlackv = blackv[row & 1];
for (; col < W - 3; col += 4) {
const vfloat linecorrv = SQRV(vmaxf(LVFU(cfablur[row * W + col]) - rowBlackv, epsv)) /
(vmaxf(LVFU(cfablur1[row * W + col]) - rowBlackv, epsv) * vmaxf(LVFU(cfablur2[row * W + col]) - rowBlackv, epsv));
const vfloat valv = LVFU(rawData[row][col]) - rowBlackv;
STVFU(rawData[row][col], valv * linecorrv + rowBlackv);
}
#endif
for (; col < W; ++col) {
const float linecorr = SQR(std::max(1e-5f, cfablur[row * W + col] - black[c4[row & 1][col & 1]])) /
(std::max(1e-5f, cfablur1[row * W + col] - black[c4[row & 1][col & 1]]) * std::max(1e-5f, cfablur2[row * W + col] - black[c4[row & 1][col & 1]]));
rawData[row][col] = (rawData[row][col] - black[c4[row & 1][col & 1]]) * linecorr + black[c4[row & 1][col & 1]];
}
}
} else if (ri->getSensorType() == ST_FUJI_XTRANS) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int row = 0; row < H; ++row) {
for (int col = 0; col < W; ++col) {
const int c = ri->XTRANSFC(row, col);
const float hlinecorr = std::max(1e-5f, cfablur[(row) * W + col] - black[c]) / std::max(1e-5f, cfablur1[(row) * W + col] - black[c]);
const float vlinecorr = std::max(1e-5f, cfablur[(row) * W + col] - black[c]) / std::max(1e-5f, cfablur2[(row) * W + col] - black[c]);
rawData[row][col] = (rawData[row][col] - black[c]) * hlinecorr * vlinecorr + black[c];
}
}
}
delete [] cfablur1;
delete [] cfablur2;
}
delete [] cfablur;
}
} /* namespace */
Reference in New Issue View Git Blame Copy Permalink