xymp
/
HDRplus
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Progress on denoising and separated functions

Browse Source 
I put the temporal and spatial denoising into 2 separate functions and am implementing finding the absolute difference. I have also edited the merge.h file to add the function signatures.
main
cvachha 3 years ago committed by Haohua-Lyu
parent 7c16c3aa81
commit 8300993446
1 changed files with 38 additions and 64 deletions
Show Stats Download Patch File Download Diff File

102
src/merge.cpp
Unescape Escape View File

@ -566,6 +566,7 @@ std::pair<double, double> merge::getNoiseParams( int ISO, \
// TODO: 4.2 Temporal Denoising
std::vector<cv::Mat> temporal_denoise(std::vector<cv::Mat> reference_tiles, std::vector<cv::Mat> reference_tiles_DFT, std::vector<float> noise_varaince) {
//goal: temporially denoise using the weiner filter
//input:
//1. array of 2D dft tiles of the reference image
@ -577,7 +578,7 @@ std::pair<double, double> merge::getNoiseParams( int ISO, \
//tile_size = TILE_SIZE;
/*
double temporal_factor = 8.0 //8 by default
double temporal_noise_scaling = (pow(TILE_SIZE,2) * (1.0/16*2))*temporal_factor;
@ -606,36 +607,53 @@ std::pair<double, double> merge::getNoiseParams( int ISO, \
//get the dft of the alternate image
//std::vector<cv::Mat> alternate_tiles_DFT;
std::vector<cv::Mat> tile_differences = reference_tiles_DFT - alternate_tiles_DFT_list;
std::vector<cv::Mat> tile_sq_asolute_diff = tile_differences; //squared absolute difference is tile_differences.real**2 + tile_differnce.imag**2; //also tile_dist
//find the squared absolute difference across all the tiles
//std::vector<cv::Mat> tile_differences = reference_tiles_DFT - alternate_tiles_DFT_list;
std::vector<cv::Mat> A = tile_sq_asolute_diff/(tile_sq_asolute_diff+noise_variance)
//find reference_tiles_DFT - alternate_tiles_DFT_list
std::vector<std::vector<cv::Mat>> tile_difference_list; //list of tile differences
for (auto individual_alternate_tile_DFT : alternate_tiles_DFT_list) {
std::vector<cv::Mat> single_tile_difference = reference_tiles_DFT - individual_alternate_tile_DFT;
tile_difference_list.push_back(single_tile_difference);
}
std::vector<cv::Mat> merged_image_tiles_fft = alternate_tiles_DFT_list + A * tile_differences;
//use merged_image_tiles_fft into part 4.3
// std::vector<cv::Mat> tile_sq_asolute_diff = tile_differences; //squared absolute difference is tile_differences.real**2 + tile_differnce.imag**2; //also tile_dist
std::vector<cv::Mat> tile_sq_asolute_diff = tile_differences; //squared absolute difference is tile_differences.real**2 + tile_differnce.imag**2; //also tile_dist
//get the real and imaginary components
/*
std::vector<std::vector<cv::Mat>> absolute_difference_list;
for (auto individual_difference : tile_difference_list) {
for (int i =0; i < individual_difference.rows; i++ ) {
std::complex<double>* row_ptr = tile_sq_asolute_diff.ptr<std::complex<double>>(i);
for (int j = 0; j< individual_difference.cols*individual_difference.channels(); j++) {
row_ptr = math.pow(individual_difference.at<std::complex<double>>(i,j).real(),2)+math.pow(individual_difference.at<std::complex<double>>(i,j).imag(),2); //.real and .imag
}
}
//std::vector<cv::Mat> single_tile_difference = individual_difference.at<std::complex<double>>(0,0).real(); //.real and .imag
absolute_difference_list.push_back(single_tile_difference);
}
*/
// TODO: 4.3 Spatial Denoising
//find the squared absolute difference across all the tiles
// Apply IFFT on reference tiles (frequency to spatial)
std::vector<cv::Mat> denoised_tiles;
for (auto dft_tile : reference_tiles_DFT) {
cv::Mat denoised_tile;
cv::dft(dft_tile, denoised_tile, cv::DFT_INVERSE | cv::DFT_REAL_OUTPUT);
denoised_tile.convertTo(denoised_tile, CV_16U);
denoised_tiles.push_back(denoised_tile);
}
reference_tiles = denoised_tiles;
//adding after here
std::vector<cv::Mat> A = tile_sq_asolute_diff/(tile_sq_asolute_diff+noise_variance)
std::vector<cv::Mat> merged_image_tiles_fft = alternate_tiles_DFT_list + A * tile_differences;
return merged_image_tiles_fft
}
std::vector<cv::Mat> spatial_denoise(std::vector<cv::Mat> reference_tiles, std::vector<cv::Mat> reference_tiles_DFT, std::vector<float> noise_varaince) {
double spatial_factor = 1; //to be added
double spatial_noise_scaling = (pow(TILE_SIZE,2) * (1.0/16*2))*spatial_factor;
@ -644,52 +662,8 @@ std::pair<double, double> merge::getNoiseParams( int ISO, \
std::vector<cv::Mat> WienerCoeff = denoised_tiles*spatial_noise_scaling*noise_variance;
merged_channel_tiles_spatial = reference_tiles*spatial_tile_dist/(spatial_tile_dist+WienerCoeff)
//apply the cosineWindow2D over the merged_channel_tiles_spatial and reconstruct the image
*/
// 4.4 Cosine Window Merging
// Process tiles through 2D cosine window
std::vector<cv::Mat> windowed_tiles;
for (auto tile : reference_tiles) {
windowed_tiles.push_back(cosineWindow2D(tile));
}
// Merge tiles
return mergeTiles(windowed_tiles, channel_image.rows, channel_image.cols);
}
//Helper function to get the channels from the input image
std::vector<ushort> getChannels(cv::Mat input_image){
std::vector<ushort> channels[4];
for (int y = 0; y < input_image.rows; ++y) {
for (int x = 0; x < input_image.cols; ++x) {
if (y % 2 == 0) {
if (x % 2 == 0) {
channels[0].push_back(input_image.at<ushort>(y, x));
}
else {
channels[1].push_back(input_image.at<ushort>(y, x));
}
}
else {
if (x % 2 == 0) {
channels[2].push_back(input_image.at<ushort>(y, x));
}
else {
channels[3].push_back(input_image.at<ushort>(y, x));
}
}
}
}
return channels;
}
//we should be getting the individual channel in the same place where we call the processChannel function with the reference channel in its arguments
} // namespace hdrplus
Write Preview
Loading…
Cancel
Save