Image denoising using multiple compressive reconstructed images

A novel image denoising technique is proposed using compressive reconstruction framework. Multiple subsets of pixels with partial overlap are selected from noisy image. The subsets of pixels are selected in such a way that every subset will have at least 20% different pixels compared to any other subset. Each subset of pixels is considered as vector of compressive samples from noisy image and a complete image is reconstructed in wavelet domain using compressive sensing reconstruction algorithm. The multiple images obtained from these subsets are merged using statistical techniques to obtain clean image. The proposed technique is tested on several images with different noise level and it was observed that proposed technique produced better denoising results compared to other denoising techniques using wavelets representation of noisy image.