rawTherapee/rtengine/filmnegativeproc.cc

/*
 *  This file is part of RawTherapee.
 *
 *  Copyright (c) 2019 Alberto Romei <aldrop8@gmail.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 <iostream>

#include "rawimage.h"
#include "rawimagesource.h"

#include "coord.h"
#include "mytime.h"
#include "opthelper.h"
#include "pixelsmap.h"
#include "procparams.h"
#include "rt_algo.h"
#include "rtengine.h"
#include "sleef.h"
//#define BENCHMARK
#include "StopWatch.h"

namespace
{
using rtengine::ST_BAYER;
using rtengine::ST_FUJI_XTRANS;
using rtengine::settings;

bool channelsAvg(
    const rtengine::RawImage* ri,
    int width,
    int height,
    const float* cblacksom,
    rtengine::Coord spotPos,
    int spotSize,
    std::array<float, 3>& avgs
)
{
    avgs = {}; // Channel averages

    if (ri->getSensorType() != ST_BAYER && ri->getSensorType() != ST_FUJI_XTRANS) {
        return false;
    }

    if (settings->verbose) {
        printf("Spot coord:  x=%d y=%d\n", spotPos.x, spotPos.y);
    }

    const int half_spot_size = spotSize / 2;

    const int& x1 = spotPos.x - half_spot_size;
    const int& x2 = spotPos.x + half_spot_size;
    const int& y1 = spotPos.y - half_spot_size;
    const int& y2 = spotPos.y + half_spot_size;

    if (x1 < 0 || x2 > width || y1 < 0 || y2 > height) {
        return false; // Spot goes outside bounds, bail out.
    }

    std::array<int, 3> pxCount = {}; // Per-channel sample counts

    for (int c = x1; c < x2; ++c) {
        for (int r = y1; r < y2; ++r) {
            const int ch = ri->getSensorType() == ST_BAYER ? ri->FC(r, c) : ri->XTRANSFC(r, c);

            ++pxCount[ch];

            // Sample the original unprocessed values from RawImage, subtracting black levels.
            // Scaling is irrelevant, as we are only interested in the ratio between two spots.
            avgs[ch] += ri->data[r][c] - cblacksom[ch];
        }
    }

    for (int ch = 0; ch < 3; ++ch) {
        avgs[ch] /= pxCount[ch];
    }

    return true;
}

void calcMedians(
    const rtengine::RawImage* ri,
    float** data,
    int x1, int y1, int x2, int y2,
    std::array<float, 3>& meds
)
{

    MyTime t1, t2, t3;
    t1.set();

    // Channel vectors to calculate medians
    std::array<std::vector<float>, 3> cvs;

    // Sample one every 5 pixels, and push the value in the appropriate channel vector.
    // Choose an odd step, not a multiple of the CFA size, to get a chance to visit each channel.
    if (ri->getSensorType() == ST_BAYER) {
        for (int row = y1; row < y2; row += 5) {
            const int c0 = ri->FC(row, x1 + 0);
            const int c1 = ri->FC(row, x1 + 5);
            int col = x1;

            for (; col < x2 - 5; col += 10) {
                cvs[c0].push_back(data[row][col]);
                cvs[c1].push_back(data[row][col + 5]);
            }

            if (col < x2) {
                cvs[c0].push_back(data[row][col]);
            }
        }
    } else if (ri->getSensorType() == ST_FUJI_XTRANS) {
        for (int row = y1; row < y2; row += 5) {
            const std::array<unsigned int, 6> cs = {
                ri->XTRANSFC(row, x1 + 0),
                ri->XTRANSFC(row, x1 + 5),
                ri->XTRANSFC(row, x1 + 10),
                ri->XTRANSFC(row, x1 + 15),
                ri->XTRANSFC(row, x1 + 20),
                ri->XTRANSFC(row, x1 + 25)
            };
            int col = x1;

            for (; col < x2 - 25; col += 30) {
                for (int c = 0; c < 6; ++c) {
                    cvs[cs[c]].push_back(data[row][col + c * 5]);
                }
            }

            for (int c = 0; col < x2; col += 5, ++c) {
                cvs[cs[c]].push_back(data[row][col]);
            }
        }
    }

    t2.set();

    if (settings->verbose) {
        printf("Median vector fill loop time us: %d\n", t2.etime(t1));
    }

    t2.set();

    for (int c = 0; c < 3; ++c) {
        // Find median values for each channel
        if (!cvs[c].empty()) {
            rtengine::findMinMaxPercentile(cvs[c].data(), cvs[c].size(), 0.5f, meds[c], 0.5f, meds[c], true);
        }
    }

    t3.set();

    if (settings->verbose) {
        printf("Sample count: R=%zu, G=%zu, B=%zu\n", cvs[0].size(), cvs[1].size(), cvs[2].size());
        printf("Median calc time us: %d\n", t3.etime(t2));
    }

}

std::array<double, 3> calcWBMults(
    const rtengine::ColorTemp& wb,
    const rtengine::ImageMatrices& imatrices,
    const rtengine::RawImage *ri,
    const float ref_pre_mul[4])
{
    std::array<double, 3> wb_mul;
    double r, g, b;
    wb.getMultipliers(r, g, b);
    wb_mul[0] = imatrices.cam_rgb[0][0] * r + imatrices.cam_rgb[0][1] * g + imatrices.cam_rgb[0][2] * b;
    wb_mul[1] = imatrices.cam_rgb[1][0] * r + imatrices.cam_rgb[1][1] * g + imatrices.cam_rgb[1][2] * b;
    wb_mul[2] = imatrices.cam_rgb[2][0] * r + imatrices.cam_rgb[2][1] * g + imatrices.cam_rgb[2][2] * b;

    for (int c = 0; c < 3; ++c) {
        wb_mul[c] = ri->get_pre_mul(c) / wb_mul[c] / ref_pre_mul[c];
    }

    // Normalize max channel gain to 1.0
    float mg = rtengine::max(wb_mul[0], wb_mul[1], wb_mul[2]);

    for (int c = 0; c < 3; ++c) {
        wb_mul[c] /= mg;
    }

    return wb_mul;
}

}

bool rtengine::RawImageSource::getFilmNegativeExponents(Coord2D spotA, Coord2D spotB, int tran, const procparams::FilmNegativeParams &currentParams, std::array<float, 3>& newExps)
{
    newExps = {
        static_cast<float>(currentParams.redRatio * currentParams.greenExp),
        static_cast<float>(currentParams.greenExp),
        static_cast<float>(currentParams.blueRatio * currentParams.greenExp)
    };

    constexpr int spotSize = 32; // TODO: Make this configurable?

    Coord spot;
    std::array<float, 3> clearVals;
    std::array<float, 3> denseVals;

    // Get channel averages in the two spots, sampling from the original ri->data buffer.
    // NOTE: rawData values might be affected by CA correction, FlatField, etc, so:
    //   rawData[y][x] == (ri->data[y][x] - cblacksom[c]) * scale_mul[c]
    // is not always true. To calculate exponents on the exact values, we should keep
    // a copy of the rawData buffer after preprocessing. Worth the memory waste?

    // Sample first spot
    transformPosition(spotA.x, spotA.y, tran, spot.x, spot.y);

    if (!channelsAvg(ri, W, H, cblacksom, spot, spotSize, clearVals)) {
        return false;
    }

    // Sample second spot
    transformPosition(spotB.x, spotB.y, tran, spot.x, spot.y);

    if (!channelsAvg(ri, W, H, cblacksom, spot, spotSize, denseVals)) {
        return false;
    }

    // Detect which one is the dense spot, based on green channel
    if (clearVals[1] < denseVals[1]) {
        std::swap(clearVals, denseVals);
    }

    if (settings->verbose) {
        printf("Clear film values: R=%g G=%g B=%g\n", static_cast<double>(clearVals[0]), static_cast<double>(clearVals[1]), static_cast<double>(clearVals[2]));
        printf("Dense film values: R=%g G=%g B=%g\n", static_cast<double>(denseVals[0]), static_cast<double>(denseVals[1]), static_cast<double>(denseVals[2]));
    }

    const float denseGreenRatio = clearVals[1] / denseVals[1];

    // Calculate logarithms in arbitrary base
    const auto logBase =
        [](float base, float num) -> float
        {
            return std::log(num) / std::log(base);
        };

    // Calculate exponents for each channel, based on the ratio between the bright and dark values,
    // compared to the ratio in the reference channel (green)
    for (int ch = 0; ch < 3; ++ch) {
        if (ch == 1) {
            newExps[ch] = 1.f;  // Green is the reference channel
        } else {
            newExps[ch] = rtengine::LIM(logBase(clearVals[ch] / denseVals[ch], denseGreenRatio), 0.3f, 4.f);
        }
    }

    if (settings->verbose) {
        printf("New exponents:  R=%g G=%g B=%g\n", static_cast<double>(newExps[0]), static_cast<double>(newExps[1]), static_cast<double>(newExps[2]));
    }

    return true;
}

bool rtengine::RawImageSource::getRawSpotValues(Coord2D spotCoord, int spotSize, int tran, const procparams::FilmNegativeParams &params, std::array<float, 3>& rawValues)
{
    Coord spot;
    transformPosition(spotCoord.x, spotCoord.y, tran, spot.x, spot.y);

    if (settings->verbose) {
        printf("Transformed coords: %d,%d\n", spot.x, spot.y);
    }

    if (spotSize < 4) {
        return false;
    }

    // Calculate averages of raw unscaled channels
    if (!channelsAvg(ri, W, H, cblacksom, spot, spotSize, rawValues)) {
        return false;
    }

    if (settings->verbose) {
        printf("Raw spot values: R=%g, G=%g, B=%g\n", rawValues[0], rawValues[1], rawValues[2]);
    }

    return true;
}

void rtengine::RawImageSource::filmNegativeProcess(const procparams::FilmNegativeParams &params, std::array<float, 3>& filmBaseValues)
{
//    BENCHFUNMICRO

    if (!params.enabled) {
        return;
    }

    // Exponents are expressed as positive in the parameters, so negate them in order
    // to get the reciprocals.
    const std::array<float, 3> exps = {
        static_cast<float>(-params.redRatio * params.greenExp),
        static_cast<float>(-params.greenExp),
        static_cast<float>(-params.blueRatio * params.greenExp)
    };

    constexpr float MAX_OUT_VALUE = 65000.f;

    // Get multipliers for a known, fixed WB setting, that will be the starting point
    // for balancing the converted image.
    const std::array<double, 3> wb_mul = calcWBMults(
            ColorTemp(3500., 1., 1., "Custom"), imatrices, ri, ref_pre_mul);


    if (rtengine::settings->verbose) {
        printf("Fixed WB mults: %g %g %g\n", wb_mul[0], wb_mul[1], wb_mul[2]);
    }

    // Compensation factor to restore the non-autoWB initialGain (see RawImageSource::load)
    const float autoGainComp = camInitialGain / initialGain;

    std::array<float, 3> mults;  // Channel normalization multipliers

    // If film base values are set in params, use those
    if (filmBaseValues[0] <= 0.f) {
        // ...otherwise, the film negative tool might have just been enabled on this image,
        // without any previous setting. So, estimate film base values from channel medians

        std::array<float, 3> medians;

        // Special value for backwards compatibility with profiles saved by RT 5.7
        const bool oldChannelScaling = filmBaseValues[0] == -1.f;

        // If using the old channel scaling method, get medians from the whole current image,
        // reading values from the already-scaled rawData buffer.
        if (oldChannelScaling) {
            calcMedians(ri, rawData, 0, 0, W, H, medians);
        } else {
            // Cut 20% border from medians calculation. It will probably contain outlier values
            // from the film holder, which will bias the median result.
            const int bW = W * 20 / 100;
            const int bH = H * 20 / 100;
            calcMedians(ri, rawData, bW, bH, W - bW, H - bH, medians);
        }

        // Un-scale rawData medians
        for (int c = 0; c < 3; ++c) {
            medians[c] /= scale_mul[c];
        }

        if (settings->verbose) {
            printf("Channel medians: R=%g, G=%g, B=%g\n", medians[0], medians[1], medians[2]);
        }

        for (int c = 0; c < 3; ++c) {

            // Estimate film base values, so that in the output data, each channel
            // median will correspond to 1/24th of MAX.
            filmBaseValues[c] = pow_F(24.f / 512.f, 1.f / exps[c]) * medians[c];

            if (oldChannelScaling) {
                // If using the old channel scaling method, apply WB multipliers here to undo their
                // effect later, since fixed wb compensation was not used in previous version.
                // Also, undo the effect of the autoGainComp factor (see below).
                filmBaseValues[c] /= pow_F((wb_mul[c] / autoGainComp), 1.f / exps[c]);// / autoGainComp;
            }

        }
    }


    // Calculate multipliers based on previously obtained film base input values.

    // Apply current scaling coefficients to raw, unscaled base values.
    std::array<float, 3> fb = {
        filmBaseValues[0] * scale_mul[0],
        filmBaseValues[1] * scale_mul[1],
        filmBaseValues[2] * scale_mul[2]
    };

    if (settings->verbose) {
        printf("Input film base values: %g %g %g\n", fb[0], fb[1], fb[2]);
    }

    for (int c = 0; c < 3; ++c) {
        // Apply channel exponents, to obtain the corresponding base values in the output data
        fb[c] = pow_F(rtengine::max(fb[c], 1.f), exps[c]);

        // Determine the channel multiplier so that the film base value is 1/512th of max.
        mults[c] = (MAX_OUT_VALUE / 512.f) / fb[c];

        // Un-apply the fixed WB multipliers, to reverse their effect later in the WB tool.
        // This way, the output image will be adapted to this fixed WB setting
        mults[c] /= wb_mul[c];

        // Also compensate for the initialGain difference between the default scaling (forceAutoWB=true)
        // and the non-autoWB scaling (see camInitialGain).
        // This effectively restores camera scaling multipliers, and gives us stable multipliers
        // (not depending on image content).
        mults[c] *= autoGainComp;

    }

    if (settings->verbose) {
        printf("Output film base values: %g %g %g\n", static_cast<double>(fb[0]), static_cast<double>(fb[1]), static_cast<double>(fb[2]));
        printf("Computed multipliers: %g %g %g\n", static_cast<double>(mults[0]), static_cast<double>(mults[1]), static_cast<double>(mults[2]));
    }


    constexpr float CLIP_VAL = 65535.f;

    MyTime t1, t2, t3;

    t1.set();

    if (ri->getSensorType() == ST_BAYER) {
#ifdef __SSE2__
        const vfloat onev = F2V(1.f);
        const vfloat clipv = F2V(CLIP_VAL);
#endif

#ifdef _OPENMP
        #pragma omp parallel for schedule(dynamic, 16)
#endif

        for (int row = 0; row < H; ++row) {
            int col = 0;
            // Avoid trouble with zeroes, minimum pixel value is 1.
            const float exps0 = exps[FC(row, col)];
            const float exps1 = exps[FC(row, col + 1)];
            const float mult0 = mults[FC(row, col)];
            const float mult1 = mults[FC(row, col + 1)];
#ifdef __SSE2__
            const vfloat expsv = _mm_setr_ps(exps0, exps1, exps0, exps1);
            const vfloat multsv = _mm_setr_ps(mult0, mult1, mult0, mult1);

            for (; col < W - 3; col += 4) {
                STVFU(rawData[row][col], vminf(multsv * pow_F(vmaxf(LVFU(rawData[row][col]), onev), expsv), clipv));
            }

#endif // __SSE2__

            for (; col < W - 1; col += 2) {
                rawData[row][col] = rtengine::min(mult0 * pow_F(rtengine::max(rawData[row][col], 1.f), exps0), CLIP_VAL);
                rawData[row][col + 1] = rtengine::min(mult1 * pow_F(rtengine::max(rawData[row][col + 1], 1.f), exps1), CLIP_VAL);
            }

            if (col < W) {
                rawData[row][col] = rtengine::min(mult0 * pow_F(rtengine::max(rawData[row][col], 1.f), exps0), CLIP_VAL);
            }
        }
    } else if (ri->getSensorType() == ST_FUJI_XTRANS) {
#ifdef __SSE2__
        const vfloat onev = F2V(1.f);
        const vfloat clipv = F2V(CLIP_VAL);
#endif

#ifdef _OPENMP
        #pragma omp parallel for schedule(dynamic, 16)
#endif

        for (int row = 0; row < H; row ++) {
            int col = 0;
            // Avoid trouble with zeroes, minimum pixel value is 1.
            const std::array<float, 6> expsc = {
                exps[ri->XTRANSFC(row, 0)],
                exps[ri->XTRANSFC(row, 1)],
                exps[ri->XTRANSFC(row, 2)],
                exps[ri->XTRANSFC(row, 3)],
                exps[ri->XTRANSFC(row, 4)],
                exps[ri->XTRANSFC(row, 5)]
            };
            const std::array<float, 6> multsc = {
                mults[ri->XTRANSFC(row, 0)],
                mults[ri->XTRANSFC(row, 1)],
                mults[ri->XTRANSFC(row, 2)],
                mults[ri->XTRANSFC(row, 3)],
                mults[ri->XTRANSFC(row, 4)],
                mults[ri->XTRANSFC(row, 5)]
            };
#ifdef __SSE2__
            const vfloat expsv0 = _mm_setr_ps(expsc[0], expsc[1], expsc[2], expsc[3]);
            const vfloat expsv1 = _mm_setr_ps(expsc[4], expsc[5], expsc[0], expsc[1]);
            const vfloat expsv2 = _mm_setr_ps(expsc[2], expsc[3], expsc[4], expsc[5]);
            const vfloat multsv0 = _mm_setr_ps(multsc[0], multsc[1], multsc[2], multsc[3]);
            const vfloat multsv1 = _mm_setr_ps(multsc[4], multsc[5], multsc[0], multsc[1]);
            const vfloat multsv2 = _mm_setr_ps(multsc[2], multsc[3], multsc[4], multsc[5]);

            for (; col < W - 11; col += 12) {
                STVFU(rawData[row][col], vminf(multsv0 * pow_F(vmaxf(LVFU(rawData[row][col]), onev), expsv0), clipv));
                STVFU(rawData[row][col + 4], vminf(multsv1 * pow_F(vmaxf(LVFU(rawData[row][col + 4]), onev), expsv1), clipv));
                STVFU(rawData[row][col + 8], vminf(multsv2 * pow_F(vmaxf(LVFU(rawData[row][col + 8]), onev), expsv2), clipv));
            }

#endif // __SSE2__

            for (; col < W - 5; col += 6) {
                for (int c = 0; c < 6; ++c) {
                    rawData[row][col + c] = rtengine::min(multsc[c] * pow_F(rtengine::max(rawData[row][col + c], 1.f), expsc[c]), CLIP_VAL);
                }
            }

            for (int c = 0; col < W; col++, c++) {
                rawData[row][col + c] = rtengine::min(multsc[c] * pow_F(rtengine::max(rawData[row][col + c], 1.f), expsc[c]), CLIP_VAL);
            }
        }
    }

    t2.set();

    if (settings->verbose) {
        printf("Pow loop time us: %d\n", t2.etime(t1));
    }

    t2.set();

    PixelsMap bitmapBads(W, H);

    int totBP = 0; // Hold count of bad pixels to correct

    if (ri->getSensorType() == ST_BAYER) {
#ifdef _OPENMP
        #pragma omp parallel for reduction(+:totBP) schedule(dynamic,16)
#endif

        for (int i = 0; i < H; ++i) {
            for (int j = 0; j < W; ++j) {
                if (rawData[i][j] >= MAX_OUT_VALUE) {
                    bitmapBads.set(j, i);
                    ++totBP;
                }
            }
        }

        if (totBP > 0) {
            interpolateBadPixelsBayer(bitmapBads, rawData);
        }

    } else if (ri->getSensorType() == ST_FUJI_XTRANS) {
#ifdef _OPENMP
        #pragma omp parallel for reduction(+:totBP) schedule(dynamic,16)
#endif

        for (int i = 0; i < H; ++i) {
            for (int j = 0; j < W; ++j) {
                if (rawData[i][j] >= MAX_OUT_VALUE) {
                    bitmapBads.set(j, i);
                    totBP++;
                }
            }
        }

        if (totBP > 0) {
            interpolateBadPixelsXtrans(bitmapBads);
        }
    }

    t3.set();

    if (settings->verbose) {
        printf("Bad pixels count: %d\n", totBP);
        printf("Bad pixels interpolation time us: %d\n", t3.etime(t2));
    }
}