rawTherapee/rtengine/ipsharpenedges.cc
2020-02-09 14:14:46 +01:00

218 lines
10 KiB
C++

/*
* This file is part of RawTherapee.
*
* Copyright (c) 2004-2020 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 <memory>
#include "improcfun.h"
#include "labimage.h"
#include "procparams.h"
#include "rt_math.h"
namespace {
#ifdef __SSE2__
bool inintervalLoRo(float a, float b, float c)
{
return a < std::max(b, c) && a > std::min(b, c);
}
float selectweight(float a, float b, float low, float high)
{
const float minVal = std::min(a,b);
const float maxVal = std::max(a,b);
const float res = (minVal < 0.45f * maxVal) ? low : high;
return (minVal > 0.05f * maxVal) ? res : high;
}
#else
bool inintervalLoRo(float a, float b, float c)
{
return (a < b && a > c) || (a < c && a > b);
}
float selectweight(float a, float b, float low, float high)
{
if ((a < 0.45f * b && a > 0.05f * b) || (b < 0.45f * a && b > 0.05f * a)) {
return low;
} else {
return high;
}
}
#endif
}
namespace rtengine
{
// To the extent possible under law, Manuel Llorens <manuelllorens@gmail.com>
// has waived all copyright and related or neighboring rights to this work.
// This work is published from: Spain.
// Thanks to Manuel for this excellent job (Jacques Desmis JDC or frej83)
void ImProcFunctions::MLsharpen (LabImage* lab)
{
// JD: this algorithm maximize clarity of images; it does not play on accutance. It can remove (partially) the effects of the AA filter)
// I think we can use this algorithm alone in most cases, or first to clarify image and if you want a very little USM (unsharp mask sharpening) after...
if (!params->sharpenEdge.enabled || params->sharpenEdge.amount == 0) {
return;
}
const int width = lab->W, height = lab->H;
constexpr float chmax[3] = {1.f / 8.f, 1.f / 3.f, 1.f / 3.f};
const int width2 = 2 * width;
constexpr float eps2 = 0.001f; //prevent divide by zero
const float amount = params->sharpenEdge.amount / 100.0;
const float amountby3 = params->sharpenEdge.amount / 300.0;
std::unique_ptr<float[]> L(new float[width * height]);
const int channels = params->sharpenEdge.threechannels ? 1 : 3;
const int passes = params->sharpenEdge.passes;
for (int c = 0; c < channels; ++c) { // c=0 Luminance only
float** channel = c == 0 ? lab->L : c == 1 ? lab->a : lab->b;
for (int p = 0; p < passes; ++p) {
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < height; ++i) {
for (int j = 0; j < width; ++j) {
L[i * width + j] = channel[i][j] / 327.68f;
}
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16)
#endif
for (int j = 2; j < height - 2; j++) {
for (int i = 2, offset = j * width + i; i < width - 2; i++, offset++) {
// weight functions
const float wH = eps2 + std::fabs(L[offset + 1] - L[offset - 1]);
const float wV = eps2 + std::fabs(L[offset + width] - L[offset - width]);
float s = 2.f / (2.f + std::fabs(wH - wV));
float wD1 = eps2 + std::fabs(L[offset + width + 1] - L[offset - width - 1]) * s;
float wD2 = eps2 + std::fabs(L[offset + width - 1] - L[offset - width + 1]) * s;
s = wD1;
wD1 /= wD2;
wD2 /= s;
const float v = L[offset];
float lumH, lumV, lumD1, lumD2;
lumH = lumV = lumD1 = lumD2 = v;
// contrast detection
const float contrast = std::min(std::sqrt(SQR(L[offset + 1] - L[offset - 1]) + SQR(L[offset + width] - L[offset - width])) * chmax[c], 1.f);
// new possible values
if (inintervalLoRo(v, L[offset - 1], L[offset + 1])) {
float f1 = std::fabs(L[offset - 2] - L[offset - 1]);
float f2 = L[offset - 1] - v;
float f3 = (L[offset - 1] - L[offset - width]) * (L[offset - 1] - L[offset + width]);
float f4 = std::sqrt(std::fabs((L[offset - 1] - L[offset - width2]) * (L[offset - 1] - L[offset + width2])));
const float difL = f1 * SQR(f2 * f3) * f4;
if (difL > 0.f) {
f1 = std::fabs(L[offset + 2] - L[offset + 1]);
f2 = L[offset + 1] - v;
f3 = (L[offset + 1] - L[offset - width]) * (L[offset + 1] - L[offset + width]);
f4 = std::sqrt(std::fabs((L[offset + 1] - L[offset - width2]) * (L[offset + 1] - L[offset + width2])));
const float difR = f1 * SQR(f2 * f3) * f4;
if (difR > 0.f) {
lumH = (L[offset - 1] * difR + L[offset + 1] * difL) / (difL + difR);
lumH = intp(contrast, lumH, v);
}
}
}
if (inintervalLoRo(v, L[offset - width], L[offset + width])) {
float f1 = std::fabs(L[offset - width2] - L[offset - width]);
float f2 = L[offset - width] - v;
float f3 = (L[offset - width] - L[offset - 1]) * (L[offset - width] - L[offset + 1]);
float f4 = std::sqrt(std::fabs((L[offset - width] - L[offset - 2]) * (L[offset - width] - L[offset + 2])));
const float difT = f1 * SQR(f2 * f3) * f4;
if (difT > 0.f) {
f1 = std::fabs(L[offset + width2] - L[offset + width]);
f2 = L[offset + width] - v;
f3 = (L[offset + width] - L[offset - 1]) * (L[offset + width] - L[offset + 1]);
f4 = std::sqrt(std::fabs((L[offset + width] - L[offset - 2]) * (L[offset + width] - L[offset + 2])));
const float difB = f1 * SQR(f2 * f3) * f4;
if (difB > 0.f) {
lumV = (L[offset - width] * difB + L[offset + width] * difT) / (difT + difB);
lumV = intp(contrast, lumV, v);
}
}
}
if (inintervalLoRo(v, L[offset - 1 - width], L[offset + 1 + width])) {
float f1 = std::fabs(L[offset - 2 - width2] - L[offset - 1 - width]);
float f2 = L[offset - 1 - width] - v;
float f3 = (L[offset - 1 - width] - L[offset - width + 1]) * (L[offset - 1 - width] - L[offset + width - 1]);
float f4 = std::sqrt(std::fabs((L[offset - 1 - width] - L[offset - width2 + 2]) * (L[offset - 1 - width] - L[offset + width2 - 2])));
const float difLT = f1 * SQR(f2 * f3) * f4;
if (difLT > 0.f) {
f1 = std::fabs(L[offset + 2 + width2] - L[offset + 1 + width]);
f2 = L[offset + 1 + width] - v;
f3 = (L[offset + 1 + width] - L[offset - width + 1]) * (L[offset + 1 + width] - L[offset + width - 1]);
f4 = std::sqrt(std::fabs((L[offset + 1 + width] - L[offset - width2 + 2]) * (L[offset + 1 + width] - L[offset + width2 - 2])));
const float difRB = f1 * SQR(f2 * f3) * f4;
if (difRB > 0.f) {
lumD1 = (L[offset - 1 - width] * difRB + L[offset + 1 + width] * difLT) / (difLT + difRB);
lumD1 = intp(contrast, lumD1, v);
}
}
}
if (inintervalLoRo(v, L[offset + 1 - width], L[offset - 1 + width])) {
float f1 = std::fabs(L[offset - 2 + width2] - L[offset - 1 + width]);
float f2 = L[offset - 1 + width] - v;
float f3 = (L[offset - 1 + width] - L[offset - width - 1]) * (L[offset - 1 + width] - L[offset + width + 1]);
float f4 = std::sqrt(std::fabs((L[offset - 1 + width] - L[offset - width2 - 2]) * (L[offset - 1 + width] - L[offset + width2 + 2])));
const float difLB = f1 * SQR(f2 * f3) * f4;
if (difLB > 0.f) {
f1 = std::fabs(L[offset + 2 - width2] - L[offset + 1 - width]);
f2 = L[offset + 1 - width] - v;
f3 = (L[offset + 1 - width] - L[offset + width + 1]) * (L[offset + 1 - width] - L[offset - width - 1]);
f4 = std::sqrt(std::fabs((L[offset + 1 - width] - L[offset + width2 + 2]) * (L[offset + 1 - width] - L[offset - width2 - 2])));
const float difRT = f1 * SQR(f2 * f3) * f4;
if (difRT > 0.f) {
lumD2 = (L[offset + 1 - width] * difLB + L[offset - 1 + width] * difRT) / (difLB + difRT);
lumD2 = intp(contrast, lumD2, v);
}
}
}
// final mix
// avoid sharpening diagonals too much
const float weight = selectweight(wH, wV, amountby3, amount);
if (c == 0) {
if (v < 92.f) {
channel[j][i] = std::fabs(327.68f * intp(weight, (lumH * wH + lumV * wV + lumD1 * wD1 + lumD2 * wD2) / (wH + wV + wD1 + wD2), v)); // fabs because lab->L always > 0
}
} else {
channel[j][i] = 327.68f * intp(weight, (lumH * wH + lumV * wV + lumD1 * wD1 + lumD2 * wD2) / (wH + wV + wD1 + wD2), v);
}
}
}
}
}
}
}