rawTherapee/rtengine/cfa_linedn_RT.cc

428 lines
14 KiB
C++

////////////////////////////////////////////////////////////////
//
// CFA line denoise by DCT filtering
//
// copyright (c) 2008-2010 Emil Martinec <ejmartin@uchicago.edu>
// parallelized 2013 by Ingo Weyrich
//
// code dated: June 7, 2010
//
// cfa_linedn_RT.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 <http://www.gnu.org/licenses/>.
//
////////////////////////////////////////////////////////////////
#include <cmath>
#include "rtengine.h"
#include "rawimagesource.h"
#include "rt_math.h"
#define TS 224 // Tile size of 224 instead of 512 speeds up processing
#define CLASS
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
using namespace std;
using namespace rtengine;
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
void RawImageSource::CLASS cfa_linedn(float noise)
{
// local variables
int height=H, width=W;
const float clip_pt = 0.8*initialGain* 65535.0;
float eps=1e-5; //tolerance to avoid dividing by zero
const float gauss[5] = {0.20416368871516755, 0.18017382291138087, 0.1238315368057753, 0.0662822452863612, 0.02763055063889883};
const float rolloff[8] = {0, 0.135335, 0.249352, 0.411112, 0.606531, 0.800737, 0.945959, 1}; //gaussian with sigma=3
const float window[8] = {0, .25, .75, 1, 1, .75, .25, 0}; //sine squared
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
if (plistener) {
plistener->setProgressStr ("Line Denoise...");
plistener->setProgress (0.0);
}
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
float noisevar=SQR(3*noise*65535); // _noise_ (as a fraction of saturation) is input to the algorithm
float noisevarm4 = 4.0f * noisevar;
volatile double progress = 0.0;
float* RawDataTmp = (float*)malloc( width*height*sizeof(float));
#pragma omp parallel
{
// allocate memory and assure the arrays don't have same 64 byte boundary to avoid L1 conflict misses
char *buffer = (char*)malloc(4 * TS * TS * sizeof(float)+ 3*64);
float *cfain = (float*)(buffer);
float *cfablur =(float*)(buffer+(TS*TS*sizeof(float))+ 1 * 64);
float *cfadiff =(float*)(buffer+(2*TS*TS*sizeof(float))+ 2 * 64);
float *cfadn =(float*)(buffer+(3*TS*TS*sizeof(float))+ 3 * 64);
float linehvar[4], linevvar[4], noisefactor[4][8][2], coeffsq;
float dctblock[4][8][8];
#pragma omp for
for(int i=0;i<height;i++)
for(int j=0;j<width;j++)
RawDataTmp[i*width+j] = rawData[i][j];
// Main algorithm: Tile loop
#pragma omp for schedule(dynamic) collapse(2)
for (int top=0; top < height-16; top += TS-32)
for (int left=0; left < width-16; left += TS-32) {
int bottom = min(top+TS,height);
int right = min(left+TS, width);
int numrows = bottom - top;
int numcols = right - left;
int indx1;
// load CFA data; data should be in linear gamma space, before white balance multipliers are applied
for (int rr=top; rr < top+numrows; rr++)
for (int cc=left, indx=(rr-top)*TS; cc < left+numcols; cc++, indx++) {
cfain[indx] = rawData[rr][cc];
}
//pad the block to a multiple of 16 on both sides
if (numcols < TS) {
indx1=numcols % 16;
for (int i=0; i<(16-indx1); i++)
for (int rr=0; rr<numrows; rr++)
cfain[(rr)*TS+numcols+i]=cfain[(rr)*TS+numcols-i-1];
numcols += 16-indx1;
}
if (numrows < TS) {
indx1=numrows % 16;
for (int i=0; i<(16-indx1); i++)
for (int cc=0; cc<numcols; cc++)
cfain[(numrows+i)*TS+cc]=cfain[(numrows-i-1)*TS+cc];
numrows += 16-indx1;
}
//The cleaning algorithm starts here
//gaussian blur of CFA data
for (int rr=8; rr < numrows-8; rr++){
for (int indx=rr*TS; indx < rr*TS+numcols; indx++) {
cfablur[indx]=gauss[0]*cfain[indx];
for (int i=1; i<5; i++) {
cfablur[indx] += gauss[i]*(cfain[indx-(2*i)*TS]+cfain[indx+(2*i)*TS]);
}
}
for (int indx=rr*TS+8; indx < rr*TS+numcols-8; indx++) {
cfadn[indx] = gauss[0]*cfablur[indx];
for (int i=1; i<5; i++) {
cfadn[indx] += gauss[i]*(cfablur[indx-2*i]+cfablur[indx+2*i]);
}
cfadiff[indx]=cfain[indx]-cfadn[indx]; // hipass cfa data
}
}
//begin block DCT
for (int rr=8; rr < numrows-22; rr+=8) // (rr,cc) shift by 8 to overlap blocks
for (int cc=8; cc < numcols-22; cc+=8) {
for (int ey=0; ey<2; ey++) // (ex,ey) specify RGGB subarray
for (int ex=0; ex<2; ex++) {
//grab an 8x8 block of a given RGGB channel
for (int i=0; i<8; i++)
for (int j=0; j<8; j++) {
dctblock[2*ey+ex][i][j]=cfadiff[(rr+2*i+ey)*TS+cc+2*j+ex];
}
ddct8x8s(-1, dctblock[2*ey+ex]); //forward DCT
}
for (int ey=0; ey<2; ey++) // (ex,ey) specify RGGB subarray
for (int ex=0; ex<2; ex++) {
linehvar[2*ey+ex]=linevvar[2*ey+ex]=0;
for (int i=4; i<8; i++) {
linehvar[2*ey+ex] += SQR(dctblock[2*ey+ex][0][i]);
linevvar[2*ey+ex] += SQR(dctblock[2*ey+ex][i][0]);
}
//Wiener filter for line denoising; roll off low frequencies
for (int i=1; i<8; i++) {
coeffsq = SQR(dctblock[2*ey+ex][i][0]);//vertical
noisefactor[2*ey+ex][i][0] = coeffsq/(coeffsq+rolloff[i]*noisevar+eps);
coeffsq = SQR(dctblock[2*ey+ex][0][i]);//horizontal
noisefactor[2*ey+ex][i][1] = coeffsq/(coeffsq+rolloff[i]*noisevar+eps);
// noisefactor labels are [RGGB subarray][row/col position][0=vert,1=hor]
}
}
//horizontal lines
if (noisevarm4>(linehvar[0]+linehvar[1])) {//horizontal lines
for (int i=1; i<8; i++) {
dctblock[0][0][i] *= 0.5f*(noisefactor[0][i][1]+noisefactor[1][i][1]);//or should we use MIN???
dctblock[1][0][i] *= 0.5f*(noisefactor[0][i][1]+noisefactor[1][i][1]);//or should we use MIN???
}
}
if (noisevarm4>(linehvar[2]+linehvar[3])) {//horizontal lines
for (int i=1; i<8; i++) {
dctblock[2][0][i] *= 0.5f*(noisefactor[2][i][1]+noisefactor[3][i][1]);//or should we use MIN???
dctblock[3][0][i] *= 0.5f*(noisefactor[2][i][1]+noisefactor[3][i][1]);//or should we use MIN???
}
}
//vertical lines
if (noisevarm4>(linevvar[0]+linevvar[2])) {//vertical lines
for (int i=1; i<8; i++) {
dctblock[0][i][0] *= 0.5f*(noisefactor[0][i][0]+noisefactor[2][i][0]);//or should we use MIN???
dctblock[2][i][0] *= 0.5f*(noisefactor[0][i][0]+noisefactor[2][i][0]);//or should we use MIN???
}
}
if (noisevarm4>(linevvar[1]+linevvar[3])) {//vertical lines
for (int i=1; i<8; i++) {
dctblock[1][i][0] *= 0.5f*(noisefactor[1][i][0]+noisefactor[3][i][0]);//or should we use MIN???
dctblock[3][i][0] *= 0.5f*(noisefactor[1][i][0]+noisefactor[3][i][0]);//or should we use MIN???
}
}
for (int ey=0; ey<2; ey++) // (ex,ey) specify RGGB subarray
for (int ex=0; ex<2; ex++) {
ddct8x8s(1, dctblock[2*ey+ex]); //inverse DCT
//multiply by window fn and add to output (cfadn)
for (int i=0; i<8; i++)
for (int j=0; j<8; j++) {
cfadn[(rr+2*i+ey)*TS+cc+2*j+ex] += window[i]*window[j]*dctblock[2*ey+ex][i][j];
}
}
}
// %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// copy smoothed results to temporary buffer
for (int rr=16; rr < numrows-16; rr++) {
int row = rr + top;
for (int col=16+left, indx=rr*TS+16; indx < rr*TS+numcols-16; indx++, col++) {
if (rawData[row][col]<clip_pt && cfadn[indx]<clip_pt)
RawDataTmp[row*width+col] = CLIP((int)(cfadn[indx]+ 0.5f));
}
}
if(plistener) {
progress+=(double)((TS-32)*(TS-32))/(height*width);
if (progress>1.0)
{
progress=1.0;
}
plistener->setProgress(progress);
}
}
// clean up
free(buffer);
// copy temporary buffer back to image matrix
#pragma omp for
for(int i=0;i<height;i++)
for(int j=0;j<width;j++)
rawData[i][j] = RawDataTmp[i*width+j];
} // end of parallel processing
free(RawDataTmp);
}
#undef TS
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
/*
Discrete Cosine Transform Code
Copyright(C) 1997 Takuya OOURA (email: ooura@mmm.t.u-tokyo.ac.jp).
You may use, copy, modify this code for any purpose and
without fee. You may distribute this ORIGINAL package.
*/
/*
Short Discrete Cosine Transform
data length :8x8
method :row-column, radix 4 FFT
functions
ddct8x8s : 8x8 DCT
function prototypes
void ddct8x8s(int isgn, float **a);
*/
/*
-------- 8x8 DCT (Discrete Cosine Transform) / Inverse of DCT --------
[definition]
<case1> Normalized 8x8 IDCT
C[k1][k2] = (1/4) * sum_j1=0^7 sum_j2=0^7
a[j1][j2] * s[j1] * s[j2] *
cos(pi*j1*(k1+1/2)/8) *
cos(pi*j2*(k2+1/2)/8), 0<=k1<8, 0<=k2<8
(s[0] = 1/sqrt(2), s[j] = 1, j > 0)
<case2> Normalized 8x8 DCT
C[k1][k2] = (1/4) * s[k1] * s[k2] * sum_j1=0^7 sum_j2=0^7
a[j1][j2] *
cos(pi*(j1+1/2)*k1/8) *
cos(pi*(j2+1/2)*k2/8), 0<=k1<8, 0<=k2<8
(s[0] = 1/sqrt(2), s[j] = 1, j > 0)
[usage]
<case1>
ddct8x8s(1, a);
<case2>
ddct8x8s(-1, a);
[parameters]
a[0...7][0...7] :input/output data (double **)
output data
a[k1][k2] = C[k1][k2], 0<=k1<8, 0<=k2<8
*/
/* Cn_kR = sqrt(2.0/n) * cos(pi/2*k/n) */
/* Cn_kI = sqrt(2.0/n) * sin(pi/2*k/n) */
/* Wn_kR = cos(pi/2*k/n) */
/* Wn_kI = sin(pi/2*k/n) */
#define C8_1R 0.49039264020161522456
#define C8_1I 0.09754516100806413392
#define C8_2R 0.46193976625564337806
#define C8_2I 0.19134171618254488586
#define C8_3R 0.41573480615127261854
#define C8_3I 0.27778511650980111237
#define C8_4R 0.35355339059327376220
#define W8_4R 0.70710678118654752440
void RawImageSource::ddct8x8s(int isgn, float a[8][8])
{
int j;
float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
float xr, xi;
if (isgn < 0) {
for (j = 0; j <= 7; j++) {
x0r = a[0][j] + a[7][j];
x1r = a[0][j] - a[7][j];
x0i = a[2][j] + a[5][j];
x1i = a[2][j] - a[5][j];
x2r = a[4][j] + a[3][j];
x3r = a[4][j] - a[3][j];
x2i = a[6][j] + a[1][j];
x3i = a[6][j] - a[1][j];
xr = x0r + x2r;
xi = x0i + x2i;
a[0][j] = C8_4R * (xr + xi);
a[4][j] = C8_4R * (xr - xi);
xr = x0r - x2r;
xi = x0i - x2i;
a[2][j] = C8_2R * xr - C8_2I * xi;
a[6][j] = C8_2R * xi + C8_2I * xr;
xr = W8_4R * (x1i - x3i);
x1i = W8_4R * (x1i + x3i);
x3i = x1i - x3r;
x1i += x3r;
x3r = x1r - xr;
x1r += xr;
a[1][j] = C8_1R * x1r - C8_1I * x1i;
a[7][j] = C8_1R * x1i + C8_1I * x1r;
a[3][j] = C8_3R * x3r - C8_3I * x3i;
a[5][j] = C8_3R * x3i + C8_3I * x3r;
}
for (j = 0; j <= 7; j++) {
x0r = a[j][0] + a[j][7];
x1r = a[j][0] - a[j][7];
x0i = a[j][2] + a[j][5];
x1i = a[j][2] - a[j][5];
x2r = a[j][4] + a[j][3];
x3r = a[j][4] - a[j][3];
x2i = a[j][6] + a[j][1];
x3i = a[j][6] - a[j][1];
xr = x0r + x2r;
xi = x0i + x2i;
a[j][0] = C8_4R * (xr + xi);
a[j][4] = C8_4R * (xr - xi);
xr = x0r - x2r;
xi = x0i - x2i;
a[j][2] = C8_2R * xr - C8_2I * xi;
a[j][6] = C8_2R * xi + C8_2I * xr;
xr = W8_4R * (x1i - x3i);
x1i = W8_4R * (x1i + x3i);
x3i = x1i - x3r;
x1i += x3r;
x3r = x1r - xr;
x1r += xr;
a[j][1] = C8_1R * x1r - C8_1I * x1i;
a[j][7] = C8_1R * x1i + C8_1I * x1r;
a[j][3] = C8_3R * x3r - C8_3I * x3i;
a[j][5] = C8_3R * x3i + C8_3I * x3r;
}
} else {
for (j = 0; j <= 7; j++) {
x1r = C8_1R * a[1][j] + C8_1I * a[7][j];
x1i = C8_1R * a[7][j] - C8_1I * a[1][j];
x3r = C8_3R * a[3][j] + C8_3I * a[5][j];
x3i = C8_3R * a[5][j] - C8_3I * a[3][j];
xr = x1r - x3r;
xi = x1i + x3i;
x1r += x3r;
x3i -= x1i;
x1i = W8_4R * (xr + xi);
x3r = W8_4R * (xr - xi);
xr = C8_2R * a[2][j] + C8_2I * a[6][j];
xi = C8_2R * a[6][j] - C8_2I * a[2][j];
x0r = C8_4R * (a[0][j] + a[4][j]);
x0i = C8_4R * (a[0][j] - a[4][j]);
x2r = x0r - xr;
x2i = x0i - xi;
x0r += xr;
x0i += xi;
a[0][j] = x0r + x1r;
a[7][j] = x0r - x1r;
a[2][j] = x0i + x1i;
a[5][j] = x0i - x1i;
a[4][j] = x2r - x3i;
a[3][j] = x2r + x3i;
a[6][j] = x2i - x3r;
a[1][j] = x2i + x3r;
}
for (j = 0; j <= 7; j++) {
x1r = C8_1R * a[j][1] + C8_1I * a[j][7];
x1i = C8_1R * a[j][7] - C8_1I * a[j][1];
x3r = C8_3R * a[j][3] + C8_3I * a[j][5];
x3i = C8_3R * a[j][5] - C8_3I * a[j][3];
xr = x1r - x3r;
xi = x1i + x3i;
x1r += x3r;
x3i -= x1i;
x1i = W8_4R * (xr + xi);
x3r = W8_4R * (xr - xi);
xr = C8_2R * a[j][2] + C8_2I * a[j][6];
xi = C8_2R * a[j][6] - C8_2I * a[j][2];
x0r = C8_4R * (a[j][0] + a[j][4]);
x0i = C8_4R * (a[j][0] - a[j][4]);
x2r = x0r - xr;
x2i = x0i - xi;
x0r += xr;
x0i += xi;
a[j][0] = x0r + x1r;
a[j][7] = x0r - x1r;
a[j][2] = x0i + x1i;
a[j][5] = x0i - x1i;
a[j][4] = x2r - x3i;
a[j][3] = x2r + x3i;
a[j][6] = x2i - x3r;
a[j][1] = x2i + x3r;
}
}
}