Removal of mixed noise by adaptive linear combination of weighted order statistics (LWOS) filters based on local statistics

It is well known that median filters are very useful for image restoration. The success of median filters is based on two properties: edge preservation and efficient impulsive noise attenuation. However, the median filter is not a perfect filtering operation. It may remove important image details and not sufficiently attenuate non-impulsive noise. The main reason is that the median filter uses only rank-order information of the input data within the filter window, and discards its original temporal-order information. In order to utilize both rank- and temporal-order information of input data, several classes of rank order based filters have been developed in recent years, such as Ll filters[l], weighted median filters[2], FIR-WOS(weighted order statistics) hybrid(FWH) filters [3]. J. Song and Y. H. Lee have also proposed a filter whose structure provide a unified framework for representing both linear FIR and median-type nonlinear filters called linear combination of weighted order statistics (LWOS) filters[4]. The LWOS filter is combination of L-filters and WOS filters.