fast_nlmeans_multi_denoising_invoker.hpp

/*M///////////////////////////////////////////////////////////////////////////////////////
//
//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
//  By downloading, copying, installing or using the software you agree to this license.
//  If you do not agree to this license, do not download, install,
//  copy or use the software.
//
//
//                        Intel License Agreement
//                For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective icvers.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
//   * Redistribution's of source code must retain the above copyright notice,
//     this list of conditions and the following disclaimer.
//
//   * Redistribution's in binary form must reproduce the above copyright notice,
//     this list of conditions and the following disclaimer in the documentation
//     and/or other materials provided with the distribution.
//
//   * The name of Intel Corporation may not be used to endorse or promote products
//     derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/

#ifndef __OPENCV_FAST_NLMEANS_MULTI_DENOISING_INVOKER_HPP__
#define __OPENCV_FAST_NLMEANS_MULTI_DENOISING_INVOKER_HPP__

#include "precomp.hpp"
#include <limits>

#include "fast_nlmeans_denoising_invoker_commons.hpp"
#include "arrays.hpp"

using namespace cv;

template <typename T, typename IT, typename UIT, typename D, typename WT>
struct FastNlMeansMultiDenoisingInvoker :
        ParallelLoopBody
{
public:
    FastNlMeansMultiDenoisingInvoker(const std::vector<Mat>& srcImgs, int imgToDenoiseIndex,
                                     int temporalWindowSize, Mat& dst, int template_window_size,
                                     int search_window_size, const float *h);

    void operator() (const Range& range) const CV_OVERRIDE;

private:
    void operator= (const FastNlMeansMultiDenoisingInvoker&);

    int rows_;
    int cols_;

    Mat& dst_;

    std::vector<Mat> extended_srcs_;
    Mat main_extended_src_;
    int border_size_;

    int template_window_size_;
    int search_window_size_;
    int temporal_window_size_;

    int template_window_half_size_;
    int search_window_half_size_;
    int temporal_window_half_size_;

    typename pixelInfo<WT>::sampleType fixed_point_mult_;
    int almost_template_window_size_sq_bin_shift;
    std::vector<WT> almost_dist2weight;

    void calcDistSumsForFirstElementInRow(int i, Array3d<int>& dist_sums,
                                          Array4d<int>& col_dist_sums,
                                          Array4d<int>& up_col_dist_sums) const;

    void calcDistSumsForElementInFirstRow(int i, int j, int first_col_num,
                                          Array3d<int>& dist_sums, Array4d<int>& col_dist_sums,
                                          Array4d<int>& up_col_dist_sums) const;
};

template <typename T, typename IT, typename UIT, typename D, typename WT>
FastNlMeansMultiDenoisingInvoker<T, IT, UIT, D, WT>::FastNlMeansMultiDenoisingInvoker(
    const std::vector<Mat>& srcImgs,
    int imgToDenoiseIndex,
    int temporalWindowSize,
    cv::Mat& dst,
    int template_window_size,
    int search_window_size,
    const float *h) :
        dst_(dst), extended_srcs_(srcImgs.size())
{
    CV_Assert(srcImgs.size() > 0);
    CV_Assert(srcImgs[0].channels() == pixelInfo<T>::channels);

    rows_ = srcImgs[0].rows;
    cols_ = srcImgs[0].cols;

    template_window_half_size_ = template_window_size / 2;
    search_window_half_size_ = search_window_size / 2;
    temporal_window_half_size_ = temporalWindowSize / 2;

    template_window_size_ = template_window_half_size_ * 2 + 1;
    search_window_size_ = search_window_half_size_ * 2 + 1;
    temporal_window_size_ = temporal_window_half_size_ * 2 + 1;

    border_size_ = search_window_half_size_ + template_window_half_size_;
    for (int i = 0; i < temporal_window_size_; i++)
        copyMakeBorder(srcImgs[imgToDenoiseIndex - temporal_window_half_size_ + i], extended_srcs_[i],
            border_size_, border_size_, border_size_, border_size_, cv::BORDER_DEFAULT);

    main_extended_src_ = extended_srcs_[temporal_window_half_size_];
    const IT max_estimate_sum_value =
        (IT)temporal_window_size_ * (IT)search_window_size_ * (IT)search_window_size_ * (IT)pixelInfo<T>::sampleMax();
    fixed_point_mult_ = (int)std::min<IT>(std::numeric_limits<IT>::max() / max_estimate_sum_value,
                                          pixelInfo<WT>::sampleMax());

    // precalc weight for every possible l2 dist between blocks
    // additional optimization of precalced weights to replace division(averaging) by binary shift
    int template_window_size_sq = template_window_size_ * template_window_size_;
    almost_template_window_size_sq_bin_shift = 0;
    while (1 << almost_template_window_size_sq_bin_shift < template_window_size_sq)
        almost_template_window_size_sq_bin_shift++;

    int almost_template_window_size_sq = 1 << almost_template_window_size_sq_bin_shift;
    double almost_dist2actual_dist_multiplier = (double) almost_template_window_size_sq / template_window_size_sq;

    int max_dist = D::template maxDist<T>();
    int almost_max_dist = (int)(max_dist / almost_dist2actual_dist_multiplier + 1);
    almost_dist2weight.resize(almost_max_dist);

    for (int almost_dist = 0; almost_dist < almost_max_dist; almost_dist++)
    {
        double dist = almost_dist * almost_dist2actual_dist_multiplier;
        almost_dist2weight[almost_dist] =
            D::template calcWeight<T, WT>(dist, h, fixed_point_mult_);
    }

    // additional optimization init end
    if (dst_.empty())
        dst_ = Mat::zeros(srcImgs[0].size(), srcImgs[0].type());
}

template <typename T, typename IT, typename UIT, typename D, typename WT>
void FastNlMeansMultiDenoisingInvoker<T, IT, UIT, D, WT>::operator() (const Range& range) const
{
    int row_from = range.start;
    int row_to = range.end - 1;

    Array3d<int> dist_sums(temporal_window_size_, search_window_size_, search_window_size_);

    // for lazy calc optimization
    Array4d<int> col_dist_sums(template_window_size_, temporal_window_size_, search_window_size_, search_window_size_);

    int first_col_num = -1;
    Array4d<int> up_col_dist_sums(cols_, temporal_window_size_, search_window_size_, search_window_size_);

    for (int i = row_from; i <= row_to; i++)
    {
        for (int j = 0; j < cols_; j++)
        {
            int search_window_y = i - search_window_half_size_;
            int search_window_x = j - search_window_half_size_;

            // calc dist_sums
            if (j == 0)
            {
                calcDistSumsForFirstElementInRow(i, dist_sums, col_dist_sums, up_col_dist_sums);
                first_col_num = 0;
            }
            else
            {
                // calc cur dist_sums using previous dist_sums
                if (i == row_from)
                {
                    calcDistSumsForElementInFirstRow(i, j, first_col_num,
                        dist_sums, col_dist_sums, up_col_dist_sums);

                }
                else
                {
                    int ay = border_size_ + i;
                    int ax = border_size_ + j + template_window_half_size_;

                    int start_by =
                        border_size_ + i - search_window_half_size_;

                    int start_bx =
                        border_size_ + j - search_window_half_size_ + template_window_half_size_;

                    T a_up = main_extended_src_.at<T>(ay - template_window_half_size_ - 1, ax);
                    T a_down = main_extended_src_.at<T>(ay + template_window_half_size_, ax);

                    // copy class member to local variable for optimization
                    int search_window_size = search_window_size_;

                    for (int d = 0; d < temporal_window_size_; d++)
                    {
                        Mat cur_extended_src = extended_srcs_[d];
                        Array2d<int> cur_dist_sums = dist_sums[d];
                        Array2d<int> cur_col_dist_sums = col_dist_sums[first_col_num][d];
                        Array2d<int> cur_up_col_dist_sums = up_col_dist_sums[j][d];
                        for (int y = 0; y < search_window_size; y++)
                        {
                            int* dist_sums_row = cur_dist_sums.row_ptr(y);

                            int* col_dist_sums_row = cur_col_dist_sums.row_ptr(y);
                            int* up_col_dist_sums_row = cur_up_col_dist_sums.row_ptr(y);

                            const T* b_up_ptr = cur_extended_src.ptr<T>(start_by - template_window_half_size_ - 1 + y);
                            const T* b_down_ptr = cur_extended_src.ptr<T>(start_by + template_window_half_size_ + y);

                            for (int x = 0; x < search_window_size; x++)
                            {
                                dist_sums_row[x] -= col_dist_sums_row[x];

                                col_dist_sums_row[x] = up_col_dist_sums_row[x] +
                                    D::template calcUpDownDist<T>(a_up, a_down, b_up_ptr[start_bx + x], b_down_ptr[start_bx + x]);

                                dist_sums_row[x] += col_dist_sums_row[x];
                                up_col_dist_sums_row[x] = col_dist_sums_row[x];
                            }
                        }
                    }
                }

                first_col_num = (first_col_num + 1) % template_window_size_;
            }

            // calc weights
            IT estimation[pixelInfo<T>::channels], weights_sum[pixelInfo<WT>::channels];
            for (size_t channel_num = 0; channel_num < pixelInfo<T>::channels; channel_num++)
                estimation[channel_num] = 0;
            for (size_t channel_num = 0; channel_num < pixelInfo<WT>::channels; channel_num++)
                weights_sum[channel_num] = 0;

            for (int d = 0; d < temporal_window_size_; d++)
            {
                const Mat& esrc_d = extended_srcs_[d];
                for (int y = 0; y < search_window_size_; y++)
                {
                    const T* cur_row_ptr = esrc_d.ptr<T>(border_size_ + search_window_y + y);

                    int* dist_sums_row = dist_sums.row_ptr(d, y);

                    for (int x = 0; x < search_window_size_; x++)
                    {
                        int almostAvgDist = dist_sums_row[x] >> almost_template_window_size_sq_bin_shift;

                        WT weight =  almost_dist2weight[almostAvgDist];
                        T p = cur_row_ptr[border_size_ + search_window_x + x];
                        incWithWeight<T, IT, WT>(estimation, weights_sum, weight, p);
                    }
                }
            }

            divByWeightsSum<IT, UIT, pixelInfo<T>::channels, pixelInfo<WT>::channels>(estimation,
                                                                                      weights_sum);
            dst_.at<T>(i,j) = saturateCastFromArray<T, IT>(estimation);
        }
    }
}

template <typename T, typename IT, typename UIT, typename D, typename WT>
inline void FastNlMeansMultiDenoisingInvoker<T, IT, UIT, D, WT>::calcDistSumsForFirstElementInRow(
        int i, Array3d<int>& dist_sums, Array4d<int>& col_dist_sums, Array4d<int>& up_col_dist_sums) const
{
    int j = 0;

    for (int d = 0; d < temporal_window_size_; d++)
    {
        Mat cur_extended_src = extended_srcs_[d];
        for (int y = 0; y < search_window_size_; y++)
            for (int x = 0; x < search_window_size_; x++)
            {
                dist_sums[d][y][x] = 0;
                for (int tx = 0; tx < template_window_size_; tx++)
                    col_dist_sums[tx][d][y][x] = 0;

                int start_y = i + y - search_window_half_size_;
                int start_x = j + x - search_window_half_size_;

                int* dist_sums_ptr = &dist_sums[d][y][x];
                int* col_dist_sums_ptr = &col_dist_sums[0][d][y][x];
                int col_dist_sums_step = col_dist_sums.step_size(0);
                for (int tx = -template_window_half_size_; tx <= template_window_half_size_; tx++)
                {
                    for (int ty = -template_window_half_size_; ty <= template_window_half_size_; ty++)
                    {
                        int dist = D::template calcDist<T>(
                                    main_extended_src_.at<T>(border_size_ + i + ty, border_size_ + j + tx),
                                    cur_extended_src.at<T>(border_size_ + start_y + ty, border_size_ + start_x + tx));

                        *dist_sums_ptr += dist;
                        *col_dist_sums_ptr += dist;
                    }
                    col_dist_sums_ptr += col_dist_sums_step;
                }

                up_col_dist_sums[j][d][y][x] = col_dist_sums[template_window_size_ - 1][d][y][x];
            }
    }
}

template <typename T, typename IT, typename UIT, typename D, typename WT>
inline void FastNlMeansMultiDenoisingInvoker<T, IT, UIT, D, WT>::calcDistSumsForElementInFirstRow(
    int i, int j, int first_col_num, Array3d<int>& dist_sums,
    Array4d<int>& col_dist_sums, Array4d<int>& up_col_dist_sums) const
{
    int ay = border_size_ + i;
    int ax = border_size_ + j + template_window_half_size_;

    int start_by = border_size_ + i - search_window_half_size_;
    int start_bx = border_size_ + j - search_window_half_size_ + template_window_half_size_;

    int new_last_col_num = first_col_num;

    for (int d = 0; d < temporal_window_size_; d++)
    {
        Mat cur_extended_src = extended_srcs_[d];
        for (int y = 0; y < search_window_size_; y++)
            for (int x = 0; x < search_window_size_; x++)
            {
                dist_sums[d][y][x] -= col_dist_sums[first_col_num][d][y][x];

                col_dist_sums[new_last_col_num][d][y][x] = 0;
                int by = start_by + y;
                int bx = start_bx + x;

                int* col_dist_sums_ptr = &col_dist_sums[new_last_col_num][d][y][x];
                for (int ty = -template_window_half_size_; ty <= template_window_half_size_; ty++)
                {
                    *col_dist_sums_ptr += D::template calcDist<T>(
                                main_extended_src_.at<T>(ay + ty, ax),
                                cur_extended_src.at<T>(by + ty, bx));
                }

                dist_sums[d][y][x] += col_dist_sums[new_last_col_num][d][y][x];

                up_col_dist_sums[j][d][y][x] = col_dist_sums[new_last_col_num][d][y][x];
            }
    }
}

#endif