Adaptive-window adaptive order-statistics LMS filter for impulse noise filtering in images

Abstract

This paper presents an adaptive filter with adaptive window size that is effective in filtering images corrupted with impulse noise and mixed noise (Gaussian, ..., impulse) with limited blurring effects. The filter is based on an adaptive window technique, the image data order statistics, and the LMS adaptive filter. It can be viewed as a three stages filter, the first stage is to identify impulses and passible edges, then use adaptive size calculation to isolate impulses and the third stage is adaptive filtering the remaining data using the adaptive LMS filter. This process of adaptive windowing reduces the filter size and causes reduction in computations which compensates for the excessive amount of comparisons made in sorting the image data. The proposed filter simultaneously removes both positive and negative impulses as compared to other techniques which deal with only one type of impulses at a time. Comparisons with other filtering illustrate its superiority in removing the mixed noise from images.