SVD Block Processing for Non-linear Image Noise Filtering

Abstract

A new algorithm for noise filtering, based on non-linear processing of image in blocks, using singular value decomposition (SVD) is presented in this paper. Noise filtering is performed in the domain of singular values and singular vectors. A priori noise variance knowledge is not required, because singular value-based noise variance estimation is performed in the first phase of the procedure. The non-linear filtering is based on eliminating changes in singular values and singular vectors caused by additional Gaussian white noise. Processing of image in smaller blocks makes SYD-based procedure computationally feasible. Experimental results have demonstrated the validity of the approach.