Fixed a bug in calcBlendFactor(..), #4551

This commit is contained in:
heckflosse
2018-05-16 16:40:26 +02:00
parent b42d9bc164
commit 528167e578

View File

@@ -23,7 +23,7 @@
#include "rt_math.h"
#include "sleef.c"
#include "opthelper.h"
#define BENCHMARK
//#define BENCHMARK
#include "StopWatch.h"
using namespace std;
@@ -32,9 +32,19 @@ float calcBlendFactor(float val, float threshold) {
// sigmoid function
// result is in ]0;1] range
// inflexion point is at (x, y) (threshold, 0.5)
return threshold == 0.f ? 1.f : 1.f / (1.f + xexpf(-8.f * ((val / threshold) * (2.f - threshold) - 1.f )));
return threshold == 0.f ? 1.f : 1.f / (1.f + xexpf(16.f - 16.f * val / threshold));
}
#ifdef __SSE2__
vfloat calcBlendFactor(vfloat valv, vfloat thresholdv) {
// sigmoid function
// result is in ]0;1] range
// inflexion point is at (x, y) (threshold, 0.5)
const vfloat onev = F2V(1.f);
const vfloat c16v = F2V(16.f);
vfloat resultv = onev / (onev + xexpf(c16v - c16v * valv / thresholdv));
return vself(vmaskf_eq(thresholdv, ZEROV), onev, resultv);
}
#endif
}
namespace rtengine
{
@@ -105,7 +115,7 @@ void ImProcFunctions::deconvsharpening (float** luminance, float** tmp, int W, i
if (sharpenParam.deconvamount < 1) {
return;
}
BENCHFUN
float *tmpI[H] ALIGNED16;
tmpI[0] = new float[W * H];
@@ -128,39 +138,63 @@ void ImProcFunctions::deconvsharpening (float** luminance, float** tmp, int W, i
tmpII[i][j] = luminance[i][j];
}
}
// calculate contrast based blend factors to reduce sharpening in regions with low contrast
const float contrastThreshold = sharpenParam.deconvdamping / 100.f;
const float contrastFactor = contrastThreshold == 0.f ? 0.f : 1.f / contrastThreshold;
float *LM = new float[W * H]; //allocation for Luminance
float *blend = new float[W * H]; //allocation for blend factor map
// upper border
for(int j = 0; j < 2; j++)
for(int i = 0, offset = j * W + i; i < W; i++, offset++) {
blend[offset] = 0.f;
}
#ifdef _OPENMP
#pragma omp parallel
#endif
{
#ifdef __SSE2__
vfloat contrastThresholdv = F2V(contrastThreshold);
vfloat scalev = F2V(0.0625f / 327.68f);
#endif
#ifdef _OPENMP
#pragma omp for schedule(dynamic,16)
#pragma omp for schedule(dynamic,16) nowait
#endif
for(int j = 0; j < H; j++)
for(int j = 2; j < H - 2; j++) {
blend[j * W] = blend[j * W + 1] = 0.f;
int i = 2;
int offset = j * W + i;
#ifdef __SSE2__
for(; i < W - 5; i += 4, offset += 4) {
vfloat contrastv = vsqrtf(SQRV(LVFU(luminance[j][i+1]) - LVFU(luminance[j][i-1])) + SQRV(LVFU(luminance[j+1][i]) - LVFU(luminance[j-1][i])) +
SQRV(LVFU(luminance[j][i+2]) - LVFU(luminance[j][i-2])) + SQRV(LVFU(luminance[j+2][i]) - LVFU(luminance[j-2][i]))) * scalev;
STVFU(blend[offset], calcBlendFactor(contrastv, contrastThresholdv));
}
#endif
for(; i < W - 2; i++, offset++) {
float contrast = sqrtf(SQR(luminance[j][i+1] - luminance[j][i-1]) + SQR(luminance[j+1][i] - luminance[j-1][i]) +
SQR(luminance[j][i+2] - luminance[j][i-2]) + SQR(luminance[j+2][i] - luminance[j-2][i])) * 0.0625f / 327.68f;
blend[offset] = calcBlendFactor(contrast, contrastThreshold);
}
blend[j * W + W - 2] = blend[j * W + W - 1] = 0.f;
}
#ifdef _OPENMP
#pragma omp single
#endif
{
// lower border
for(int j = H - 2; j < H; j++)
for(int i = 0, offset = j * W + i; i < W; i++, offset++) {
LM[offset] = 0.f; // adjust to [0;100] and to RT variables
}
#ifdef _OPENMP
#pragma omp for schedule(dynamic,16)
#endif
for(int j = 2; j < H - 2; j++)
for(int i = 2, offset = j * W + i; i < W - 2; i++, offset++) {
float contrast;
contrast = sqrtf(SQR(luminance[j][i+1] / 327.68f - luminance[j][i-1] / 327.68f) + SQR(luminance[j+1][i] / 327.68f - luminance[j-1][i] / 327.68f)
+ SQR(luminance[j][i+1] / 327.68f - luminance[j][i-1] / 327.68f) + SQR(luminance[j+1][i] / 327.68f - luminance[j-1][i] / 327.68f)) * 0.0625f; //for 5x5
contrast = std::min(contrast, 1.f);
LM[offset] = 1.f - calcBlendFactor(contrast, contrastThreshold);
blend[offset] = 0.f;
}
}
}
float damping = sharpenParam.deconvdamping / 5.0;
bool needdamp = false; //sharpenParam.deconvdamping > 0;
double sigma = sharpenParam.deconvradius / scale;
@@ -200,10 +234,10 @@ void ImProcFunctions::deconvsharpening (float** luminance, float** tmp, int W, i
for (int i = 0; i < H; i++)
for (int j = 0, offset = i * W + j; j < W; j++, offset++) {
luminance[i][j] = intp(LM[offset], tmpII[i][j], luminance[i][j] * p1 + max(tmpI[i][j], 0.0f) * p2);
luminance[i][j] = intp(blend[offset], luminance[i][j] * p1 + max(tmpI[i][j], 0.0f) * p2, tmpII[i][j]);
}
} // end parallel
delete [] LM;
delete [] blend;
delete [] tmpI[0];
delete [] tmpII[0];
@@ -630,7 +664,6 @@ void ImProcFunctions::MLmicrocontrast(float** luminance, int W, int H)
BENCHFUN
const int k = params->sharpenMicro.matrix ? 1 : 2;
const float contrastThreshold = params->sharpenMicro.contrast / 100;
const float contrastFactor = contrastThreshold == 0.f ? 0.f : 1.f / contrastThreshold;
// k=2 matrix 5x5 k=1 matrix 3x3
const int width = W, height = H;
const float uniform = params->sharpenMicro.uniformity; //between 0 to 100
@@ -705,7 +738,7 @@ BENCHFUN
+ SQR(LM[offset + 2] - LM[offset - 2]) + SQR(LM[offset + 2 * width] - LM[offset - 2 * width])) * 0.0625f; //for 5x5
contrast = std::min(contrast, 1.f);
float blend = 1.f - calcBlendFactor(contrast, contrastThreshold);
float blend = calcBlendFactor(contrast, contrastThreshold);
//matrix 5x5
float temp = v + 4.f *( v * (s + sqrt2 * s)); //begin 3x3
@@ -804,7 +837,7 @@ BENCHFUN
} else {
temp = 0.f;
}
luminance[j][i] = intp(blend, luminance[j][i], luminance[j][i] * (temp * temp2 + 1.f));
luminance[j][i] = intp(blend, luminance[j][i] * (temp * temp2 + 1.f), luminance[j][i]);
} else {
float temp4 = LM[offset] / tempL; //
@@ -855,7 +888,7 @@ BENCHFUN
} else {
temp = 0.f;
}
luminance[j][i] = intp(blend, luminance[j][i], luminance[j][i] / (temp * temp4 + 1.f));
luminance[j][i] = intp(blend, luminance[j][i] / (temp * temp4 + 1.f), luminance[j][i]);
}
}
}