rawTherapee/rtengine/pixelshift.cc
2024-06-26 15:37:36 -04:00

1042 lines
44 KiB
C++

////////////////////////////////////////////////////////////////
//
// Algorithm for Pentax/Sony Pixel Shift raw files with motion detection
//
// Copyright (C) 2016 - 2018 Ingo Weyrich <heckflosse67@gmx.de>
//
//
// pixelshift.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 <stack>
#include "array2D.h"
#include "gauss.h"
#include "median.h"
#include "procparams.h"
#include "rawimagesource.h"
#include "sleef.h"
#include "../rtgui/multilangmgr.h"
#include "../rtgui/options.h"
//#define BENCHMARK
#include "StopWatch.h"
namespace
{
unsigned fc(const unsigned int cfa[2][2], int r, int c) {
return cfa[r & 1][c & 1];
}
float greenDiff(float a, float b, float stddevFactor, float eperIso, float nreadIso, float prnu)
{
// calculate the difference between two green samples
float gDiff = a - b;
gDiff *= eperIso;
gDiff *= gDiff;
float avg = (a + b) * 0.5f;
avg *= eperIso;
prnu *= avg;
float stddev = stddevFactor * (avg + nreadIso + prnu * prnu);
return gDiff - stddev;
}
float nonGreenDiffCross(float right, float left, float top, float bottom, float centre, float clippedVal, float stddevFactor, float eperIso, float nreadIso, float prnu)
{
if(rtengine::max(right, left, top, bottom, centre) > clippedVal) {
return 0.f;
}
// check non green cross
float hDiff = (right + left) * 0.5f - centre;
hDiff *= eperIso;
hDiff *= hDiff;
float vDiff = (top + bottom) * 0.5f - centre;
vDiff *= eperIso;
vDiff *= vDiff;
float avg = ((right + left) + (top + bottom)) * 0.25f;
avg *= eperIso;
prnu *= avg;
float stddev = stddevFactor * (avg + nreadIso + prnu * prnu);
return std::min(hDiff, vDiff) - stddev;
}
void paintMotionMask(int index, bool showMotion, float *maskDest, float *nonMaskDest0, float *nonMaskDest1)
{
if(showMotion) {
// if showMotion is enabled colourize the pixel
maskDest[index] = 13500.f;
nonMaskDest1[index] = nonMaskDest0[index] = 0.f;
}
}
void invertMask(int xStart, int xEnd, int yStart, int yEnd, const array2D<uint8_t> &maskIn, array2D<uint8_t> &maskOut)
{
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = yStart; i < yEnd; ++i) {
#ifdef _OPENMP
#pragma omp simd
#endif
for(int j = xStart; j < xEnd; ++j) {
maskOut[i][j] = ~maskIn[i][j];
}
}
}
void xorMasks(int xStart, int xEnd, int yStart, int yEnd, const array2D<uint8_t> &maskIn, array2D<uint8_t> &maskOut)
{
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = yStart; i < yEnd; ++i) {
#ifdef _OPENMP
#pragma omp simd
#endif
for(int j = xStart; j < xEnd; ++j) {
maskOut[i][j] ^= maskIn[i][j];
}
}
}
void floodFill4Impl(int yin, int xin, int xStart, int xEnd, int yStart, int yEnd, array2D<uint8_t> &mask, std::stack<std::pair<uint16_t, uint16_t>, std::vector<std::pair<uint16_t, uint16_t>>> &coordStack)
{
coordStack.emplace(xin, yin);
while(!coordStack.empty()) {
auto coord = coordStack.top();
coordStack.pop();
auto x = coord.first, y = coord.second;
if(mask[y][x] == 255) {
auto yUp = y - 1, yDown = y + 1;
bool lastXUp = false, lastXDown = false, firstXUp = false, firstXDown = false;
mask[y][x] = 0;
if(yUp >= yStart && mask[yUp][x] == 255) {
coordStack.emplace(x, yUp);
firstXUp = lastXUp = true;
}
if(yDown < yEnd && mask[yDown][x] == 255) {
coordStack.emplace(x, yDown);
firstXDown = lastXDown = true;
}
auto xr = x + 1;
while(xr < xEnd && mask[y][xr] == 255) {
mask[y][xr] = 0;
if(yUp >= yStart && mask[yUp][xr] == 255) {
if(!lastXUp) {
coordStack.emplace(xr, yUp);
lastXUp = true;
}
} else {
lastXUp = false;
}
if(yDown < yEnd && mask[yDown][xr] == 255) {
if(!lastXDown) {
coordStack.emplace(xr, yDown);
lastXDown = true;
}
} else {
lastXDown = false;
}
xr++;
}
auto xl = x - 1;
lastXUp = firstXUp;
lastXDown = firstXDown;
while(xl >= xStart && mask[y][xl] == 255) {
mask[y][xl] = 0;
if(yUp >= yStart && mask[yUp][xl] == 255) {
if(!lastXUp) {
coordStack.emplace(xl, yUp);
lastXUp = true;
}
} else {
lastXUp = false;
}
if(yDown < yEnd && mask[yDown][xl] == 255) {
if(!lastXDown) {
coordStack.emplace(xl, yDown);
lastXDown = true;
}
} else {
lastXDown = false;
}
xl--;
}
mask[y][x] = 0;
}
}
}
void floodFill4(int xStart, int xEnd, int yStart, int yEnd, array2D<uint8_t> &mask)
{
#ifdef _OPENMP
#pragma omp parallel
#endif
{
std::stack<std::pair<uint16_t, uint16_t>, std::vector<std::pair<uint16_t, uint16_t>>> coordStack;
#ifdef _OPENMP
#pragma omp for schedule(dynamic,128) nowait
#endif
for(uint16_t i = yStart; i < yEnd; i++)
{
floodFill4Impl(i, xStart, xStart, xEnd, yStart, yEnd, mask, coordStack);
}
#ifdef _OPENMP
#pragma omp for schedule(dynamic,128) nowait
#endif
for(int16_t i = yEnd - 1; i >= 0 ; i--)
{
floodFill4Impl(i, xEnd - 1, xStart, xEnd, yStart, yEnd, mask, coordStack);
}
#ifdef _OPENMP
#pragma omp sections nowait
#endif
{
#ifdef _OPENMP
#pragma omp section
#endif
{
uint16_t i = yStart;
for(uint16_t j = xStart; j < xEnd; j++)
{
floodFill4Impl(i, j, xStart, xEnd, yStart, yEnd, mask, coordStack);
}
}
#ifdef _OPENMP
#pragma omp section
#endif
{
uint16_t i = yStart;
for(uint16_t j = xEnd - 1; j >= xStart; j--)
{
floodFill4Impl(i, j, xStart, xEnd, yStart, yEnd, mask, coordStack);
}
}
#ifdef _OPENMP
#pragma omp section
#endif
{
uint16_t i = yEnd;
for(uint16_t j = xStart; j < xEnd; j++)
{
floodFill4Impl(i, j, xStart, xEnd, yStart, yEnd, mask, coordStack);
}
}
#ifdef _OPENMP
#pragma omp section
#endif
{
uint16_t i = yEnd;
for(uint16_t j = xEnd - 1; j >= xStart; j--)
{
floodFill4Impl(i, j, xStart, xEnd, yStart, yEnd, mask, coordStack);
}
}
}
}
}
void calcFrameBrightnessFactor(unsigned int frame, uint32_t datalen, LUTu *histo[4], float brightnessFactor[4])
{
float medians[4];
for (int i = 0; i < 4; ++i) {
//find median of histogram
uint32_t lmedian = 0, count = 0;
while (count < datalen / 2) {
count += (*histo[i])[lmedian];
++lmedian;
}
const float weight = (count - datalen / 2.f) / (*histo[i])[lmedian - 1];
medians[i] = rtengine::intp(weight, (float)(lmedian - 2), (float)(lmedian - 1));
}
for (int i = 0; i < 4; ++i) {
brightnessFactor[i] = medians[frame] / medians[i];
}
}
}
using namespace std;
using namespace rtengine;
void RawImageSource::pixelshift(int winx, int winy, int winw, int winh, const procparams::RAWParams &rawParamsIn, unsigned int frame, const std::string &make, const std::string &model, float rawWpCorrection)
{
BENCHFUN
if(numFrames != 4) { // fallback for non pixelshift files
amaze_demosaic_RT(winx, winy, winw, winh, rawData, red, green, blue, options.chunkSizeAMAZE, options.measure);
return;
}
procparams::RAWParams::BayerSensor bayerParams = rawParamsIn.bayersensor;
bool motionDetection = true;
if(bayerParams.pixelShiftMotionCorrectionMethod == procparams::RAWParams::BayerSensor::PSMotionCorrectionMethod::AUTO) {
bool pixelShiftEqualBright = bayerParams.pixelShiftEqualBright;
bayerParams.setPixelShiftDefaults();
bayerParams.pixelShiftEqualBright = pixelShiftEqualBright;
} else if(bayerParams.pixelShiftMotionCorrectionMethod == procparams::RAWParams::BayerSensor::PSMotionCorrectionMethod::OFF) {
motionDetection = false;
bayerParams.pixelShiftShowMotion = false;
}
const unsigned int cfarray[2][2] = {{FC(0,0), FC(0,1)}, {FC(1,0), FC(1,1)}};
const bool showMotion = bayerParams.pixelShiftShowMotion;
const bool showOnlyMask = bayerParams.pixelShiftShowMotionMaskOnly && showMotion;
const float smoothFactor = 1.0 - bayerParams.pixelShiftSmoothFactor;
if(motionDetection) {
if(!showOnlyMask) {
if(bayerParams.pixelShiftMedian || bayerParams.pixelShiftAverage) { // We need the demosaiced frames for motion correction
if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::LMMSE)) {
lmmse_interpolate_omp(winw, winh, *(rawDataFrames[0]), red, green, blue, bayerParams.lmmse_iterations);
} else if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::AMAZEVNG4)) {
dual_demosaic_RT (true, rawParamsIn, winw, winh, *(rawDataFrames[0]), red, green, blue, bayerParams.dualDemosaicContrast, true);
} else if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::RCDVNG4)) {
dual_demosaic_RT (true, rawParamsIn, winw, winh, *(rawDataFrames[0]), red, green, blue, bayerParams.dualDemosaicContrast, true);
} else {
amaze_demosaic_RT(winx, winy, winw, winh, *(rawDataFrames[0]), red, green, blue, options.chunkSizeAMAZE, options.measure);
}
multi_array2D<float, 3> redTmp(winw, winh);
multi_array2D<float, 3> greenTmp(winw, winh);
multi_array2D<float, 3> blueTmp(winw, winh);
for(int i = 0; i < 3; i++) {
if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::LMMSE)) {
lmmse_interpolate_omp(winw, winh, *(rawDataFrames[i + 1]), redTmp[i], greenTmp[i], blueTmp[i], bayerParams.lmmse_iterations);
} else if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::AMAZEVNG4)) {
dual_demosaic_RT (true, rawParamsIn, winw, winh, *(rawDataFrames[i + 1]), redTmp[i], greenTmp[i], blueTmp[i], bayerParams.dualDemosaicContrast, true);
} else if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::RCDVNG4)) {
dual_demosaic_RT (true, rawParamsIn, winw, winh, *(rawDataFrames[i + 1]), redTmp[i], greenTmp[i], blueTmp[i], bayerParams.dualDemosaicContrast, true);
} else {
amaze_demosaic_RT(winx, winy, winw, winh, *(rawDataFrames[i + 1]), redTmp[i], greenTmp[i], blueTmp[i], options.chunkSizeAMAZE, options.measure);
}
}
if(bayerParams.pixelShiftMedian) {
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = winy + border; i < winh - border; i++) {
for(int j = winx + border; j < winw - border; j++) {
red[i][j] = median(red[i][j], redTmp[0][i + 1][j], redTmp[1][i + 1][j + 1], redTmp[2][i][j + 1]);
}
for(int j = winx + border; j < winw - border; j++) {
green[i][j] = median(green[i][j], greenTmp[0][i + 1][j], greenTmp[1][i + 1][j + 1], greenTmp[2][i][j + 1]);
}
for(int j = winx + border; j < winw - border; j++) {
blue[i][j] = median(blue[i][j], blueTmp[0][i + 1][j], blueTmp[1][i + 1][j + 1], blueTmp[2][i][j + 1]);
}
}
} else {
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = winy + border; i < winh - border; i++) {
for(int j = winx + border; j < winw - border; j++) {
red[i][j] = 0.25f * ((red[i][j] + redTmp[0][i + 1][j]) + (redTmp[1][i + 1][j + 1] + redTmp[2][i][j + 1]));
}
for(int j = winx + border; j < winw - border; j++) {
green[i][j] = 0.25f * ((green[i][j] + greenTmp[0][i + 1][j]) + (greenTmp[1][i + 1][j + 1] + greenTmp[2][i][j + 1]));
}
for(int j = winx + border; j < winw - border; j++) {
blue[i][j] = 0.25f * ((blue[i][j] + blueTmp[0][i + 1][j]) + (blueTmp[1][i + 1][j + 1] + blueTmp[2][i][j + 1]));
}
}
}
} else {
if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::LMMSE)) {
lmmse_interpolate_omp(winw, winh, rawData, red, green, blue, bayerParams.lmmse_iterations);
} else if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::AMAZEVNG4)) {
procparams::RAWParams rawParamsTmp = rawParamsIn;
rawParamsTmp.bayersensor.method = procparams::RAWParams::BayerSensor::getMethodString(procparams::RAWParams::BayerSensor::Method::AMAZEVNG4);
dual_demosaic_RT (true, rawParamsTmp, winw, winh, rawData, red, green, blue, bayerParams.dualDemosaicContrast, true);
} else if (bayerParams.pixelShiftDemosaicMethod == bayerParams.getPSDemosaicMethodString(procparams::RAWParams::BayerSensor::PSDemosaicMethod::RCDVNG4)) {
procparams::RAWParams rawParamsTmp = rawParamsIn;
rawParamsTmp.bayersensor.method = procparams::RAWParams::BayerSensor::getMethodString(procparams::RAWParams::BayerSensor::Method::RCDVNG4);
dual_demosaic_RT (true, rawParamsTmp, winw, winh, rawData, red, green, blue, bayerParams.dualDemosaicContrast, true);
} else {
amaze_demosaic_RT(winx, winy, winw, winh, rawData, red, green, blue, options.chunkSizeAMAZE, options.measure);
}
}
}
}
constexpr float stddevFactorGreen = 25.f;
constexpr float stddevFactorRed = 25.f;
constexpr float stddevFactorBlue = 25.f;
constexpr float prnu = 0.01f;
constexpr float redBlueWeight = 0.7f + 1.f;
float eperIso = bayerParams.pixelShiftEperIso;
const bool checkNonGreenCross = bayerParams.pixelShiftNonGreenCross;
const bool checkGreen = bayerParams.pixelShiftGreen;
constexpr float greenWeight = 2.f;
const bool blurMap = bayerParams.pixelShiftBlur;
const float sigma = bayerParams.pixelShiftSigma;
constexpr float noMotion = 0.99f;
constexpr float threshold = 3.f + 9 * noMotion;
const bool holeFill = bayerParams.pixelShiftHoleFill;
const bool equalBrightness = bayerParams.pixelShiftEqualBright;
const bool equalChannel = bayerParams.pixelShiftEqualBrightChannel;
const bool smoothTransitions = blurMap && bayerParams.pixelShiftSmoothFactor > 0.;
static const float nReadK3II[] = { 3.4f, // ISO 100
3.1f, // ISO 125
2.5f, // ISO 160
2.5f, // ISO 200
2.5f, // ISO 250
2.5f, // ISO 320
2.3f, // ISO 400
2.5f, // ISO 500
2.3f, // ISO 640
2.3f, // ISO 800
2.4f, // ISO 1000
2.3f, // ISO 1250
1.75f, // ISO 1600
1.75f, // ISO 2000
1.75f, // ISO 2500
1.75f, // ISO 3200
1.75f, // ISO 4000
1.75f, // ISO 5000
1.75f, // ISO 6400
1.75f, // ISO 8000
1.75f, // ISO 10000
1.5f, // ISO 12800
1.5f, // ISO 16000
1.5f, // ISO 20000
1.5f, // ISO 25600
1.5f, // ISO 32000
1.5f, // ISO 40000
1.5f // ISO 51200
};
static const float ePerIsoK3II = 0.35f;
// currently we use the same values for K-1 and K-1 Mark II, though for the K-1 Mark II the values seem a bit aggressive
static const float nReadK1[] = { 3.45f, // ISO 100
3.15f, // ISO 125
3.45f, // ISO 160
3.0f, // ISO 200
3.0f, // ISO 250
3.0f, // ISO 320
2.7f, // ISO 400
2.7f, // ISO 500
2.7f, // ISO 640
2.5f, // ISO 800
2.5f, // ISO 1000
2.5f, // ISO 1250
2.4f, // ISO 1600
2.4f, // ISO 2000
2.4f, // ISO 2500
2.4f, // ISO 3200
2.4f, // ISO 4000
2.4f, // ISO 5000
2.4f, // ISO 6400
2.4f, // ISO 8000
2.4f, // ISO 10000
2.4f, // ISO 12800
2.4f, // ISO 16000
2.4f, // ISO 20000
2.4f, // ISO 25600
2.4f, // ISO 32000
2.4f, // ISO 40000
2.4f, // ISO 51200
2.4f, // ISO 64000
2.4f, // ISO 80000
2.4f, // ISO 102400
2.4f, // ISO 128000
2.4f, // ISO 160000
2.4f, // ISO 204800
2.4f, // ISO 256000 // these are for K-1 Mark II to avoid crashes when using high-ISO files
2.4f, // ISO 320000
2.4f, // ISO 409600
2.4f, // ISO 512000
2.4f, // ISO 640000
2.4f // ISO 819200
};
static const float ePerIsoK1 = 0.75f;
// currently nReadK70 is used for K-70 and KP
static const float nReadK70[] = { 4.0f, // ISO 100
4.0f, // ISO 125
4.0f, // ISO 160
4.0f, // ISO 200
4.0f, // ISO 250
4.0f, // ISO 320
4.0f, // ISO 400
4.0f, // ISO 500
4.0f, // ISO 640
3.0f, // ISO 800
3.0f, // ISO 1000
3.0f, // ISO 1250
3.0f, // ISO 1600
3.0f, // ISO 2000
3.0f, // ISO 2500
3.0f, // ISO 3200
3.0f, // ISO 4000
3.0f, // ISO 5000
3.0f, // ISO 6400
3.0f, // ISO 8000
3.0f, // ISO 10000
3.0f, // ISO 12800
3.0f, // ISO 16000
3.0f, // ISO 20000
3.0f, // ISO 25600
3.0f, // ISO 32000
3.0f, // ISO 40000
3.0f, // ISO 51200
3.0f, // ISO 64000
3.0f, // ISO 80000
3.0f, // ISO 102400
3.0f, // ISO 128000
3.0f, // ISO 160000
3.0f, // ISO 204800
3.0f, // ISO 256000
3.0f, // ISO 320000
3.0f, // ISO 409600
3.0f, // ISO 512000
3.0f, // ISO 640000
3.0f // ISO 819200
};
static const float ePerIsoK70 = 0.5f;
// preliminary ILCE-7RM3 data, good fidelity except from A) small inaccuracy at places
// due to integer scaling quantization, B) much different noise behaviour of PDAF pixels
static const float nReadILCE7RM3[] = { 4.2f, // ISO 100
3.9f, // ISO 125
3.6f, // ISO 160
3.55f, // ISO 200
3.5f, // ISO 250
3.45f, // ISO 320
3.35f, // ISO 400
3.3f, // ISO 500
1.3f, // ISO 640
1.2f, // ISO 800
1.2f, // ISO 1000
1.2f, // ISO 1250
1.15f, // ISO 1600
1.2f, // ISO 2000
1.15f, // ISO 2500
1.15f, // ISO 3200
1.1f, // ISO 4000
1.1f, // ISO 5000
1.05f, // ISO 6400
1.05f, // ISO 8000
1.05f, // ISO 10000
1.0f, // ISO 12800
1.0f, // ISO 16000
1.0f, // ISO 20000
1.0f, // ISO 25600
1.0f, // ISO 32000
1.0f, // ISO 40000
1.0f, // ISO 51200
1.1f, // ISO 64000
1.1f, // ISO 80000
1.1f, // ISO 102400
};
static const float ePerIsoILCE7RM3 = 0.8f;
//TODO: Add data for ILCE-7RM4, and ILCE-1
if(plistener) {
plistener->setProgressStr(Glib::ustring::compose(M("TP_RAW_DMETHOD_PROGRESSBAR"), M("TP_RAW_PIXELSHIFT")));
plistener->setProgress(0.0);
}
if(motionDetection && blurMap && smoothFactor == 0.f && !showMotion) {
if(plistener) {
plistener->setProgress(1.0);
}
return;
}
float nRead;
float eperIsoModel;
int nReadIndex = static_cast<int>(round(log2(idata->getISOSpeed() / 100.f) * 3.f));
if(make.find("Sony") != string::npos) {
if(model.find("ILCE-7RM3") != string::npos) {
nRead = nReadILCE7RM3[nReadIndex];
eperIsoModel = ePerIsoILCE7RM3;
/* TODO: When we have data for missing ILCE-7RM4, and ILCE-1, add it here
} else if(model.find("ILCE-7RM4") != string::npos) {
} else if(model.find("ILCE-1") != string::npos) {
*/
} else { // default to ILCE-7RM3 for Sony cameras without data
nRead = nReadILCE7RM3[nReadIndex];
eperIsoModel = ePerIsoILCE7RM3;
}
} else { // Pentax
if(model.find("K-3") != string::npos) {
nRead = nReadK3II[nReadIndex];
eperIsoModel = ePerIsoK3II;
} else if(model.find("K-1") != string::npos) { // this also matches K-1 Mark II
nRead = nReadK1[nReadIndex];
eperIsoModel = ePerIsoK1;
} else { // as long as we don't have values for Pentax KP, we use the values from K-70
nRead = nReadK70[nReadIndex];
eperIsoModel = ePerIsoK70;
}
}
eperIsoModel *= pow(2.f, eperIso - 1.f);
eperIso = eperIsoModel * (100.f / (rawWpCorrection * idata->getISOSpeed()));
const float eperIsoRed = (eperIso / scale_mul[0]) * (65535.f / (c_white[0] - c_black[0]));
const float eperIsoGreen = (eperIso / scale_mul[1]) * (65535.f / (c_white[1] - c_black[1]));
const float eperIsoBlue = (eperIso / scale_mul[2]) * (65535.f / (c_white[2] - c_black[2]));
const float clippedRed = 65535.f / scale_mul[0];
const float clippedBlue = 65535.f / scale_mul[2];
nRead *= nRead;
// calculate channel median brightness for each frame
float greenBrightness[4] = {1.f, 1.f, 1.f, 1.f};
float redBrightness[4] = {1.f, 1.f, 1.f, 1.f};
float blueBrightness[4] = {1.f, 1.f, 1.f, 1.f};
if(equalBrightness) {
if(rawDirty) {
LUTu *histogreen[4];
LUTu *histored[4];
LUTu *histoblue[4];
for(int i = 0; i < 4; ++i) {
histogreen[i] = new LUTu(65536, LUT_CLIP_BELOW | LUT_CLIP_ABOVE, true);
histored[i] = new LUTu(65536, LUT_CLIP_BELOW | LUT_CLIP_ABOVE, true);
histoblue[i] = new LUTu(65536, LUT_CLIP_BELOW | LUT_CLIP_ABOVE, true);
}
#ifdef _OPENMP
#pragma omp parallel
#endif
{
LUTu *histogreenThr[4];
LUTu *historedThr[4];
LUTu *histoblueThr[4];
for(int i = 0; i < 4; ++i) {
histogreenThr[i] = new LUTu(65536, LUT_CLIP_BELOW | LUT_CLIP_ABOVE, true);
historedThr[i] = new LUTu(65536, LUT_CLIP_BELOW | LUT_CLIP_ABOVE, true);
histoblueThr[i] = new LUTu(65536, LUT_CLIP_BELOW | LUT_CLIP_ABOVE, true);
}
#ifdef _OPENMP
#pragma omp for schedule(dynamic,16) nowait
#endif
for(int i = winy + 1; i < winh - 1; ++i) {
int j = winx + 1;
int c = fc(cfarray, i, j);
bool bluerow = (c + fc(cfarray, i, j + 1)) == 3;
for(int j = winx + 1, offset = fc(cfarray, i, j) & 1; j < winw - 1; ++j, offset ^= 1) {
(*histogreenThr[1 - offset])[(*rawDataFrames[1 - offset])[i - offset + 1][j]]++;
(*histogreenThr[3 - offset])[(*rawDataFrames[3 - offset])[i + offset][j + 1]]++;
if(bluerow) {
(*historedThr[2 - offset])[(*rawDataFrames[2 - offset])[i + 1][j - offset + 1]]++;
(*histoblueThr[(offset << 1) + offset])[(*rawDataFrames[(offset << 1) + offset])[i][j + offset]]++;
} else {
(*historedThr[(offset << 1) + offset])[(*rawDataFrames[(offset << 1) + offset])[i][j + offset]]++;
(*histoblueThr[2 - offset])[(*rawDataFrames[2 - offset])[i + 1][j - offset + 1]]++;
}
}
}
#ifdef _OPENMP
#pragma omp critical
#endif
{
for(int i = 0; i < 4; ++i) {
(*histogreen[i]) += (*histogreenThr[i]);
delete histogreenThr[i];
(*histored[i]) += (*historedThr[i]);
delete historedThr[i];
(*histoblue[i]) += (*histoblueThr[i]);
delete histoblueThr[i];
}
}
}
calcFrameBrightnessFactor(frame, (winh - 2) * (winw - 2) / 4, histored, redBrightness);
calcFrameBrightnessFactor(frame, (winh - 2) * (winw - 2) / 4, histoblue, blueBrightness);
calcFrameBrightnessFactor(frame, (winh - 2) * (winw - 2) / 2, histogreen, greenBrightness);
for(int i = 0; i < 4; ++i) {
psRedBrightness[i] = redBrightness[i];
psGreenBrightness[i] = greenBrightness[i];
psBlueBrightness[i] = blueBrightness[i];
}
rawDirty = false;
for(int i = 0; i < 4; ++i) {
delete histored[i];
delete histoblue[i];
delete histogreen[i];
}
if(plistener) {
plistener->setProgress(0.15);
}
} else {
for(int i = 0; i < 4; ++i) {
redBrightness[i] = psRedBrightness[i];
greenBrightness[i] = psGreenBrightness[i];
blueBrightness[i] = psBlueBrightness[i];
}
}
if(!equalChannel) {
for(int i = 0; i < 4; ++i) {
redBrightness[i] = blueBrightness[i] = greenBrightness[i];
}
}
}
if(motionDetection) {
// fill channels psRed and psBlue
array2D<float> psRed(winw + 32, winh); // increase width to avoid cache conflicts
array2D<float> psBlue(winw + 32, winh);
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = winy + 1; i < winh - 1; ++i) {
float *nonGreenDest0 = psRed[i];
float *nonGreenDest1 = psBlue[i];
float ngbright[2][4] = {{redBrightness[0], redBrightness[1], redBrightness[2], redBrightness[3]},
{blueBrightness[0], blueBrightness[1], blueBrightness[2], blueBrightness[3]}
};
int ng = 0;
int j = winx + 1;
int c = fc(cfarray, i, j);
if((c + fc(cfarray, i, j + 1)) == 3) {
// row with blue pixels => swap destination pointers for non green pixels
std::swap(nonGreenDest0, nonGreenDest1);
ng ^= 1;
}
// offset to keep the code short. It changes its value between 0 and 1 for each iteration of the loop
unsigned int offset = c & 1;
for(; j < winw - 1; ++j) {
// store the non green values from the 4 frames into 2 temporary planes
nonGreenDest0[j] = (*rawDataFrames[(offset << 1) + offset])[i][j + offset] * ngbright[ng][(offset << 1) + offset];
nonGreenDest1[j] = (*rawDataFrames[2 - offset])[i + 1][j - offset + 1] * ngbright[ng ^ 1][2 - offset];
offset ^= 1; // 0 => 1 or 1 => 0
}
}
if(plistener) {
plistener->setProgress(0.3);
}
// now that the temporary planes are filled for easy access we do the motion detection
array2D<float> psMask(winw, winh);
int offsX = 0, offsY = 0;
if(!bayerParams.pixelShiftMedian) {
// We have to adjust the offsets for the selected subframe we use for areas with motion
switch(frame) {
case 0:
offsX = offsY = 0;
break;
case 1:
offsX = 0;
offsY = 1;
break;
case 2:
offsX = offsY = 1;
break;
case 3:
offsX = 1;
offsY = 0;
}
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = winy + border - offsY; i < winh - (border + offsY); ++i) {
// offset to keep the code short. It changes its value between 0 and 1 for each iteration of the loop
unsigned int offset = fc(cfarray, i, winx + border - offsX) & 1;
for(int j = winx + border - offsX; j < winw - (border + offsX); ++j, offset ^= 1) {
psMask[i][j] = noMotion;
if(checkGreen) {
if(greenDiff((*rawDataFrames[1 - offset])[i - offset + 1][j] * greenBrightness[1 - offset], (*rawDataFrames[3 - offset])[i + offset][j + 1] * greenBrightness[3 - offset], stddevFactorGreen, eperIsoGreen, nRead, prnu) > 0.f) {
psMask[i][j] = greenWeight;
// do not set the motion pixel values. They have already been set by demosaicer
continue;
}
}
if(checkNonGreenCross) {
// check red cross
float redTop = psRed[i - 1][j];
float redLeft = psRed[i][j - 1];
float redCentre = psRed[i][j];
float redRight = psRed[i][j + 1];
float redBottom = psRed[i + 1][j];
float redDiff = nonGreenDiffCross(redRight, redLeft, redTop, redBottom, redCentre, clippedRed, stddevFactorRed, eperIsoRed, nRead, prnu);
if(redDiff > 0.f) {
psMask[i][j] = redBlueWeight;
continue;
}
// check blue cross
float blueTop = psBlue[i - 1][j];
float blueLeft = psBlue[i][j - 1];
float blueCentre = psBlue[i][j];
float blueRight = psBlue[i][j + 1];
float blueBottom = psBlue[i + 1][j];
float blueDiff = nonGreenDiffCross(blueRight, blueLeft, blueTop, blueBottom, blueCentre, clippedBlue, stddevFactorBlue, eperIsoBlue, nRead, prnu);
if(blueDiff > 0.f) {
psMask[i][j] = redBlueWeight;
continue;
}
}
}
}
if(plistener) {
plistener->setProgress(0.45);
}
if(blurMap) {
#ifdef _OPENMP
#pragma omp parallel
#endif
{
gaussianBlur(psMask, psMask, winw, winh, sigma);
}
if(plistener) {
plistener->setProgress(0.6);
}
}
array2D<uint8_t> mask(winw, winh, ARRAY2D_CLEAR_DATA);
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = winy + border - offsY; i < winh - (border + offsY); ++i) {
int j = winx + border - offsX;
float v3sum[3] = {0.f};
for(int v = -1; v <= 1; v++) {
for(int h = -1; h < 1; h++) {
v3sum[1 + h] += psMask[i + v][j + h];
}
}
float blocksum = v3sum[0] + v3sum[1];
for(int voffset = 2; j < winw - (border + offsX); ++j, ++voffset) {
float colSum = psMask[i - 1][j + 1] + psMask[i][j + 1] + psMask[i + 1][j + 1];
voffset = voffset == 3 ? 0 : voffset; // faster than voffset %= 3;
blocksum -= v3sum[voffset];
blocksum += colSum;
v3sum[voffset] = colSum;
if(blocksum >= threshold) {
mask[i][j] = 255;
}
}
}
if(plistener) {
plistener->setProgress(0.75);
}
if(holeFill) {
array2D<uint8_t> maskInv(winw, winh);
invertMask(winx + border - offsX, winw - (border + offsX), winy + border - offsY, winh - (border + offsY), mask, maskInv);
floodFill4(winx + border - offsX, winw - (border + offsX), winy + border - offsY, winh - (border + offsY), maskInv);
xorMasks(winx + border - offsX, winw - (border + offsX), winy + border - offsY, winh - (border + offsY), maskInv, mask);
}
if(plistener) {
plistener->setProgress(0.9);
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = winy + border - offsY; i < winh - (border + offsY); ++i) {
#ifdef __SSE2__
// pow() is expensive => pre calculate blend factor using SSE
if(smoothTransitions) { //
vfloat onev = F2V(1.f);
vfloat smoothv = F2V(smoothFactor);
int j = winx + border - offsX;
for(; j < winw - (border + offsX) - 3; j += 4) {
vfloat blendv = vmaxf(LVFU(psMask[i][j]), onev) - onev;
blendv = pow_F(blendv, smoothv);
blendv = vself(vmaskf_eq(smoothv, ZEROV), onev, blendv);
STVFU(psMask[i][j], blendv);
}
for(; j < winw - (border + offsX); ++j) {
psMask[i][j] = smoothFactor == 0.f ? 1.f : pow_F(std::max(psMask[i][j] - 1.f, 0.f), smoothFactor);
}
}
#endif
float *greenDest = green[i + offsY];
float *redDest = red[i + offsY];
float *blueDest = blue[i + offsY];
// offset to keep the code short. It changes its value between 0 and 1 for each iteration of the loop
unsigned int offset = fc(cfarray, i, winx + border - offsX) & 1;
for(int j = winx + border - offsX; j < winw - (border + offsX); ++j, offset ^= 1) {
if(showOnlyMask) {
if(smoothTransitions) { // we want only motion mask => paint areas according to their motion (dark = no motion, bright = motion)
#ifdef __SSE2__
// use pre calculated blend factor
const float blend = psMask[i][j];
#else
const float blend = smoothFactor == 0.f ? 1.f : pow_F(std::max(psMask[i][j] - 1.f, 0.f), smoothFactor);
#endif
redDest[j + offsX] = greenDest[j + offsX] = blueDest[j + offsX] = blend * 32768.f;
} else {
redDest[j + offsX] = greenDest[j + offsX] = blueDest[j + offsX] = mask[i][j] == 255 ? 65535.f : 0.f;
}
} else if(mask[i][j] == 255) {
paintMotionMask(j + offsX, showMotion, greenDest, redDest, blueDest);
} else {
if(smoothTransitions) {
#ifdef __SSE2__
// use pre calculated blend factor
const float blend = psMask[i][j];
#else
const float blend = smoothFactor == 0.f ? 1.f : pow_F(std::max(psMask[i][j] - 1.f, 0.f), smoothFactor);
#endif
redDest[j + offsX] = intp(blend, showMotion ? 0.f : redDest[j + offsX], psRed[i][j] );
greenDest[j + offsX] = intp(blend, showMotion ? 13500.f : greenDest[j + offsX], ((*rawDataFrames[1 - offset])[i - offset + 1][j] * greenBrightness[1 - offset] + (*rawDataFrames[3 - offset])[i + offset][j + 1] * greenBrightness[3 - offset]) * 0.5f);
blueDest[j + offsX] = intp(blend, showMotion ? 0.f : blueDest[j + offsX], psBlue[i][j]);
} else {
redDest[j + offsX] = psRed[i][j];
greenDest[j + offsX] = ((*rawDataFrames[1 - offset])[i - offset + 1][j] * greenBrightness[1 - offset] + (*rawDataFrames[3 - offset])[i + offset][j + 1] * greenBrightness[3 - offset]) * 0.5f;
blueDest[j + offsX] = psBlue[i][j];
}
}
}
}
} else {
// motion detection off => combine the 4 raw frames
float ngbright[2][4] = {{redBrightness[0], redBrightness[1], redBrightness[2], redBrightness[3]},
{blueBrightness[0], blueBrightness[1], blueBrightness[2], blueBrightness[3]}
};
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for(int i = winy + 1; i < winh - 1; ++i) {
float *nonGreenDest0 = red[i];
float *nonGreenDest1 = blue[i];
int ng = 0;
int j = winx + 1;
int c = fc(cfarray, i, j);
if((c + fc(cfarray, i, j + 1)) == 3) {
// row with blue pixels => swap destination pointers for non green pixels
std::swap(nonGreenDest0, nonGreenDest1);
ng ^= 1;
}
// offset to keep the code short. It changes its value between 0 and 1 for each iteration of the loop
unsigned int offset = c & 1;
for(; j < winw - 1; ++j) {
// set red, green and blue values
green[i][j] = ((*rawDataFrames[1 - offset])[i - offset + 1][j] * greenBrightness[1 - offset] + (*rawDataFrames[3 - offset])[i + offset][j + 1] * greenBrightness[3 - offset]) * 0.5f;
nonGreenDest0[j] = (*rawDataFrames[(offset << 1) + offset])[i][j + offset] * ngbright[ng][(offset << 1) + offset];
nonGreenDest1[j] = (*rawDataFrames[2 - offset])[i + 1][j - offset + 1] * ngbright[ng ^ 1][2 - offset];
offset ^= 1; // 0 => 1 or 1 => 0
}
}
}
if(plistener) {
plistener->setProgress(1.0);
}
}