BgConvolutionImageSource.cpp

Go to the documentation of this file.

 
/*
 * BgConvolutionImageSource.cpp
 *
 *  Created on: Jun 12, 2019
 *      Author: Alejandro Alvarez
 *      Refactored out from: BackgroundConvolution.h
 */
 
#include "SEImplementation/Segmentation/BgConvolutionImageSource.h"
#include "SEFramework/Image/FunctionalImage.h"
 
namespace SourceXtractor {
 
 
BgConvolutionImageSource::BgConvolutionImageSource(std::shared_ptr<Image<DetectionImage::PixelType>> image,
                                                   std::shared_ptr<DetectionImage> variance, SeFloat threshold,
                                                   std::shared_ptr<VectorImage<SeFloat>> kernel)
  : ProcessingImageSource<DetectionImage::PixelType>(image),
    m_variance(variance), m_threshold(threshold) {
  m_kernel = VectorImage<SeFloat>::create(MirrorImage<SeFloat>::create(kernel));
}
 
std::string BgConvolutionImageSource::getRepr() const {
  return "BgConvolutionImageSource(" + getImageRepr() + ")";
}
 
static std::tuple<float, float>
applyKernel(const VectorImage<SeFloat>& kernel, ImageChunk<SeFloat>& image_chunk, ImageChunk<SeFloat>& variance_chunk,
            int start_x, int start_y, int clip_w, int clip_h, SeFloat threshold) {
  DetectionImage::PixelType total = 0.;
  DetectionImage::PixelType conv_weight = 0.;
 
  for (int cy = 0; cy < kernel.getHeight(); ++cy) {
    for (int cx = 0; cx < kernel.getWidth(); ++cx) {
      int clip_x2 = start_x + cx;
      int clip_y2 = start_y + cy;
 
      if (clip_x2 >= 0 && clip_x2 < clip_w && clip_y2 >= 0 && clip_y2 < clip_h) {
        if (variance_chunk.getValue(clip_x2, clip_y2) < threshold) {
          total += image_chunk.getValue(clip_x2, clip_y2) * kernel.getValue(cx, cy);
          conv_weight += kernel.getValue(cx, cy);
        }
      }
    }
  }
 
  return std::make_pair(total, conv_weight);
}
 
void BgConvolutionImageSource::generateTile(const std::shared_ptr<Image<DetectionImage::PixelType>>& image,
                                            ImageTileWithType<DetectionImage::PixelType>& tile, int start_x,
                                            int start_y, int width, int height) const {
  const int hx = m_kernel->getWidth() / 2;
  const int hy = m_kernel->getHeight() / 2;
  const int clip_x = std::max(start_x - hx, 0);
  const int clip_y = std::max(start_y - hy, 0);
  const int clip_w = std::min(width + hx * 2, image->getWidth() - clip_x);
  const int clip_h = std::min(height + hy * 2, image->getHeight() - clip_y);
 
  // "Materialize" the image and variance
  auto image_chunk = image->getChunk(clip_x, clip_y, clip_w, clip_h);
  auto variance_chunk = m_variance->getChunk(clip_x, clip_y, clip_w, clip_h);
 
  // Convolve both and copy out to the tile
  const int off_x = start_x - clip_x;
  const int off_y = start_y - clip_y;
  auto& tile_image = *tile.getImage();
  for (int iy = 0; iy < height; ++iy) {
    for (int ix = 0; ix < width; ++ix) {
      if (variance_chunk->getValue(ix + off_x, iy + off_y) < m_threshold) {
        DetectionImage::PixelType total = 0.;
        DetectionImage::PixelType conv_weight = 0.;
 
        std::tie(total, conv_weight) = applyKernel(*m_kernel, *image_chunk, *variance_chunk,
                                                   ix - hx + off_x, iy - hy + off_y, clip_w, clip_h,
                                                   m_threshold);
 
        // Note that because of the conditional, at least the center pixel is below the threshold,
        // so checking for conv_weight > 0 is redundant
        tile_image.setValue(ix, iy, total / conv_weight);
      }
      else {
        tile_image.setValue(ix, iy, 0.);
      }
    }
  }
}
 
 
} // end namespace SourceXtractor