Image denoising using self wavelet dictionary training

Natural images can be sparse or compressible on an appropriate basis or dictionary. Images could be denoised by assuming that the noiseless version of the image has a sparse representation on some dictionaries. Choosing a proper dictionary is significant to the performance of image denoising results. However, fixed dictionaries are limited by their ability to sparsify the images. Optimization of a dictionary could make the representation of the image they are designed to process as sparse as possible. In this paper, we introduce a framework of image denoising for the optimization of the dictionary from the noisy image itself. We show that this kind of optimization outperforms both the use of fixed dictionary and those dictionaries that are optimized independently of noisy image.