Click here for full text:
A method for using marginal statistics for image denoising
Bergman, Ruth; Hel-Or, Yacov; Nachlieli, Hila; Ruckenstein, Gitit
Keyword(s): image denoising; co-occurrence matrix; salt and pepper noise; image statistics; universal denoiser
Abstract: This invention is concerned with a method for removing noise (denoising) from gray level images using statistics that model the image textures. The denoising method presented here uses the framework of the general denoising algorithm published in . The algorithm in , known as the Discrete Universal DEnoising algorithm (DUDE), has been proved to be an optimal algorithm for data denoising in certain settings. However, when the input is a noisy gray- level image of a finite size, the sufficient conditions for optimality are not satisfied. A major difficulty in implementing DUDE for gray-level images, as opposed to binary images, lays in the high space complexity required for collecting block statistics. Furthermore, since these statistics are very sparse they do not estimate the corresponding probabilities very well. We suggest here a method to approximate the image block statistics required for DUDE by using a gray level co-occurrence matrix. Such a matrix has been used, for example in , to model the statistics of textured images.
Back to Index