The main idea of this problem is to denoise images with Gaussian noise input, using the non-local-mean algorithm. The non-local means filtering takes a mean of all pixels in the image, weighted by how similar these pixels are to the target pixel.
The denoise() function is called N * N times in main () where each time it accepts as an argument a pixel of the image with Gaussian noise and gradually through the weights w calculates and returns the corresponding pixel that will have the denoised image. The "patch (int i, int j, ...)" function returns the square neighborhood centered on the "pixel (i, j)" of the image.
In the parallel implementation the complexity from N^4 drops to N^2 through the functions global void CudaPatch() and global void nonLocalMeans(). These functions replace the for loops in the main where the patching and denoising were done.
