image_normalization.cpp

Go to the documentation of this file.

// *****************************************************************************
//
// Copyright (c) 2014, Southwest Research Institute® (SwRI®)
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//     * Redistributions of source code must retain the above copyright
//       notice, this list of conditions and the following disclaimer.
//     * Redistributions 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.
//     * Neither the name of Southwest Research Institute® (SwRI®) nor the
//       names of its contributors may 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 <COPYRIGHT HOLDER> 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.
//
// *****************************************************************************
/*
 * image_normalization.cpp
 *
 *  Created on: Jan 19, 2012
 *      Author: kkozak
 */

#include <swri_image_util/image_normalization.h>

namespace swri_image_util
{
  void normalize_illumination(cv::Mat NormImage,
                              cv::Mat SourceImage,
                              cv::Mat& DestImage)
  {
    // note that Norm image should be of type CV_32FC1 and scaled between 0 and
    // 1 (but not exactly equal to 0 anywhere)
    cv::Mat tempIm1;
    SourceImage.convertTo(tempIm1, CV_32FC1, 1.0, 0.0);
    cv::divide(tempIm1, NormImage, tempIm1, 1.0);
    cv::Mat firstImage;
    cv::Mat secondImage;
    tempIm1.convertTo(DestImage, CV_8UC1);
  }

  cv::Mat generate_normalization_image(const std::vector<cv::Mat>& image_list)
  {
    cv::Mat norm_image;
    if (image_list.empty())
    {
      return cv::Mat();
    }

    const cv::Mat& rep_im = image_list.front();

    cv::Mat image_sum(cv::Size(rep_im.cols, rep_im.rows), CV_64F);

    for (uint32_t i = 0; i < image_list.size(); i++)
    {
      cv::Mat temp_im;
      image_list[i].convertTo(temp_im, CV_64F, 1.0, 0.0);

      image_sum = image_sum + temp_im;
    }

    image_sum = image_sum / static_cast<double>(image_list.size());

    cv::Mat mean_image;
    image_sum.convertTo(mean_image, CV_8U);
    cv::Mat temp_norm_image;

    cv::medianBlur(mean_image, temp_norm_image, 45);

    cv::Mat temp_norm_image2;
    temp_norm_image.convertTo(temp_norm_image2, CV_32F);
    double max1 = 0;
    for (int32_t i = 0; i < temp_norm_image2.rows; i++)
    {
      for (int32_t j = 0; j < temp_norm_image2.cols; j++)
      {
        if (temp_norm_image2.at<float>(i, j) > max1)
        {
          max1 = temp_norm_image2.at<float>(i, j);
        }
      }
    }

    temp_norm_image2 = temp_norm_image2 * (255.0 / max1);

    cv::Mat temp_norm_image3;
    temp_norm_image2.convertTo(temp_norm_image3, CV_8U, 1.0, 0.0);


    cv::GaussianBlur(temp_norm_image3,
                     norm_image,
                     cv::Size(25, 25),
                     5,
                     5);

    return norm_image;
  }

  void MaskedBoxFilter(cv::Mat& mat, cv::Mat& mask, int32_t kernel_size)
  {
    mask.setTo(1, mask);
    cv::Mat local_sums;
    cv::boxFilter(
      mat,
      local_sums,
      mat.depth(),
      cv::Size(kernel_size, kernel_size),
      cv::Point(-1, -1),
      false);
    cv::Mat local_mask_sums;
    cv::boxFilter(
      mask,
      local_mask_sums,
      mat.depth(),
      cv::Size(kernel_size, kernel_size),
      cv::Point(-1, -1),
      false);
    cv::Mat blurred;
    cv::divide(local_sums, local_mask_sums, blurred, 1, mat.type());
    blurred.copyTo(mat, local_mask_sums != 0);
  }

  void ContrastStretch(
    int32_t grid_size,
    const cv::Mat& source_image,
    cv::Mat& dest_image,
    const cv::Mat& mask,
    double max_min,
    double min_max)
  {
    double cell_width = static_cast<double>(source_image.cols) / grid_size;
    double cell_height = static_cast<double>(source_image.rows) / grid_size;

    cv::Mat max_vals(grid_size, grid_size, CV_32F);
    cv::Mat min_vals(grid_size, grid_size, CV_32F);
    cv::Mat grid_mask = cv::Mat::ones(grid_size, grid_size, CV_8U) * 255;

    bool has_mask = !mask.empty();
    for(int i = 0; i < grid_size; i++)
    {
      for(int j = 0; j < grid_size; j++)
      {
        double minVal = 0;
        double maxVal = 0;

        int x = j * cell_width;
        int y = i * cell_height;
        int x2 = (j + 1) * cell_width;
        int y2 = (i + 1) * cell_height;
        int w = x2 - x;
        int h = y2 - y;

        cv::Rect roi = cv::Rect(x, y, w, h);
        roi.width = std::min(source_image.cols - roi.x, roi.width);
        roi.height = std::min(source_image.rows - roi.y, roi.height);

        if (has_mask)
        {
          if (cv::countNonZero(mask(roi)) > 0)
          {
            cv::minMaxLoc(source_image(roi), &minVal, &maxVal, 0, 0, mask(roi));
          }
          else
          {
            grid_mask.at<uint8_t>(i, j) = 0;
          }
        }
        else
        {
          cv::minMaxLoc(source_image(roi), &minVal, &maxVal, 0, 0);
        }

        max_vals.at<float>(i, j) = maxVal;
        min_vals.at<float>(i, j) = minVal;
      }
    }

    MaskedBoxFilter(max_vals, grid_mask, 3);
    MaskedBoxFilter(min_vals, grid_mask, 3);

    // Stretch contrast accordingly
    for(int i = 0; i < source_image.rows; i++)
    {
      int cell_y = std::min(grid_size - 1, static_cast<int>(i / cell_height));
      int cell_y_start = cell_y * cell_height;
      int cell_y_end = std::min(source_image.rows - 1, static_cast<int>((cell_y + 1) * cell_height));
      double py = (i - cell_y_start) / static_cast<double>(cell_y_end - cell_y_start);

      for(int j = 0; j < source_image.cols; j++)
      {
        // Find relevant min and max values
        int cell_x = std::min(grid_size - 1, static_cast<int>(j / cell_width));
        int cell_x_start = cell_x * cell_width;
        int cell_x_end = std::min(source_image.cols - 1, static_cast<int>((cell_x + 1) * cell_width));
        double px = (j - cell_x_start) / static_cast<double>(cell_x_end - cell_x_start);

        int cell_x1 = cell_x;
        int cell_x2 = cell_x;
        double px1 = 0.5;
        double px2 = 0.5;
        if (px < 0.5 && cell_x > 0)
        {
          cell_x1 = cell_x - 1;
          cell_x2 = cell_x;
          px2 = px + 0.5;
          px1 = 1.0 - px2;
        }
        else if (px > 0.5 && cell_x + 1 < grid_size)
        {
          cell_x1 = cell_x;
          cell_x2 = cell_x + 1;
          px1 = 1.5 - px;
          px2 = 1.0 - px1;
        }

        int cell_y1 = cell_y;
        int cell_y2 = cell_y;
        double py1 = 0.5;
        double py2 = 0.5;
        if (py < 0.5 && cell_y > 0)
        {
          cell_y1 = cell_y - 1;
          cell_y2 = cell_y;
          py2 = py + 0.5;
          py1 = 1.0 - py2;
        }
        else if (py > 0.5 && cell_y + 1 < grid_size)
        {
          cell_y1 = cell_y;
          cell_y2 = cell_y + 1;
          py1 = 1.5 - py;
          py2 = 1.0 - py1;
        }

        // Stretch histogram
        double min_x1_y1 = min_vals.at<float>(cell_y1, cell_x1);
        double max_x1_y1 = max_vals.at<float>(cell_y1, cell_x1);

        double min_x2_y1 = min_vals.at<float>(cell_y1, cell_x2);
        double max_x2_y1 = max_vals.at<float>(cell_y1, cell_x2);

        double min_x1_y2 = min_vals.at<float>(cell_y2, cell_x1);
        double max_x1_y2 = max_vals.at<float>(cell_y2, cell_x1);

        double min_x2_y2 = min_vals.at<float>(cell_y2, cell_x2);
        double max_x2_y2 = max_vals.at<float>(cell_y2, cell_x2);

        //4-point interpolation
        double xM1 = max_x1_y1 * px1 + max_x2_y1 * px2;
        double xM2 = max_x1_y2 * px1 + max_x2_y2 * px2;
        double M = xM1 * py1 + xM2 * py2;

        double xm1 = min_x1_y1 * px1 + min_x2_y1 * px2;
        double xm2 = min_x1_y2 * px1 + min_x2_y2 * px2;
        double m = xm1 * py1 + xm2 * py2;
        double minVal = m;
        double maxVal = M;

        if(maxVal > 255) maxVal = 255;
        if(minVal < 0) minVal = 0;

        // Put a bound on maxVal and minVal
        maxVal = std::max(maxVal, min_max);
        minVal = std::min(minVal, max_min);

        double val = source_image.at<uint8_t>(i,j);
        val = 255.0 * (val - minVal) / (maxVal - minVal);
        if(val > 255) val = 255;
        if(val < 0) val = 0;
        dest_image.at<uint8_t>(i,j) = val;
      }
    }

    dest_image.setTo(0, mask == 0);
  }

  void NormalizeResponse(
      const cv::Mat& src,
      cv::Mat& dst,
      int winsize,
      int ftzero,
      uchar* buf)
  {
    if (dst.empty())
    {
      dst.create(src.size(), CV_8U);
    }

    // Source code from OpenCV: modules/calib3d/src/stereobm.cpp
    int x, y, wsz2 = winsize / 2;
    int* vsum = reinterpret_cast<int*>(cv::alignPtr(buf + (wsz2 + 1) * sizeof(vsum[0]), 32));
    int scale_g = winsize * winsize / 8, scale_s = (1024 + scale_g) / (scale_g * 2);
    const int OFS = 256 * 5, TABSZ = OFS * 2 + 256;
    uchar tab[TABSZ];
    const uchar* sptr = src.ptr();
    int srcstep = src.step;
    cv::Size size = src.size();

    scale_g *= scale_s;

    for (x = 0; x < TABSZ; x++)
    {
      tab[x] = (uchar)(x - OFS < -ftzero ? 0 : x - OFS > ftzero ? ftzero * 2 : x - OFS + ftzero);
    }

    for (x = 0; x < size.width; x++)
    {
      vsum[x] = (ushort)(sptr[x] * (wsz2 + 2));
    }

    for (y = 1; y < wsz2; y++)
    {
      for (x = 0; x < size.width; x++)
      {
        vsum[x] = (ushort)(vsum[x] + sptr[srcstep*y + x]);
      }
    }

    for (y = 0; y < size.height; y++)
    {
      const uchar* top = sptr + srcstep * MAX(y - wsz2 - 1, 0);
      const uchar* bottom = sptr + srcstep * MIN(y + wsz2, size.height - 1);
      const uchar* prev = sptr + srcstep * MAX(y - 1, 0);
      const uchar* curr = sptr + srcstep * y;
      const uchar* next = sptr + srcstep * MIN(y + 1, size.height - 1);
      uchar* dptr = dst.ptr<uchar>(y);

      for (x = 0; x < size.width; x++)
      {
        vsum[x] = (ushort)(vsum[x] + bottom[x] - top[x]);
      }

      for (x = 0; x <= wsz2; x++)
      {
        vsum[-x-1] = vsum[0];
        vsum[size.width + x] = vsum[size.width - 1];
      }

      int sum = vsum[0] * (wsz2 + 1);
      for( x = 1; x <= wsz2; x++ )
      {
        sum += vsum[x];
      }

      int val = ((curr[0] * 5 + curr[1] + prev[0] + next[0]) * scale_g - sum * scale_s) >> 10;
      dptr[0] = tab[val + OFS];

      for( x = 1; x < size.width-1; x++ )
      {
        sum += vsum[x + wsz2] - vsum[x - wsz2 - 1];
        val = ((curr[x] * 4 + curr[x - 1] + curr[x + 1] + prev[x] + next[x]) * scale_g - sum * scale_s) >> 10;
        dptr[x] = tab[val + OFS];
      }

      sum += vsum[x+wsz2] - vsum[x - wsz2 - 1];
      val = ((curr[x] * 5 + curr[x - 1] + prev[x] + next[x]) * scale_g - sum * scale_s) >> 10;
      dptr[x] = tab[val + OFS];
    }
  }

  cv::Mat scale_2_8bit(const cv::Mat& image)
       {
              if (image.type() == CV_8UC1)
                return image;
              cv::Mat Image8Bit(image.rows, image.cols, CV_8U), Image8BitColor; //Define an 8bit image
              //Convert the image to 32bit float
              cv::Mat ImageFloat;
              image.convertTo(ImageFloat, CV_32F, 1, 0);
              double maxVal; //Define the max value of image
              cv::minMaxLoc(ImageFloat.reshape(1,1), NULL, &maxVal);//Extract the max value of image
              //ReScale the image to 0 to 255
              ImageFloat = ImageFloat*((1 << 8)/pow(2, ceil(log(maxVal)/log(2))));
              ImageFloat.convertTo(Image8Bit,CV_8U, 1, 0);
              return Image8Bit;
       }

       cv::Mat scale_2_8bit_color(const cv::Mat& image)
       {
              if (image.type() == CV_8UC3)
                return image;
              cv::Mat Image8Bit = scale_2_8bit(image), Image8BitColor;
              cvtColor(Image8Bit, Image8BitColor, cv::COLOR_GRAY2BGR);
              return Image8BitColor;
       }
}