Improving the filtering performance of the BDND algorithm in impulse noise removal

Switching median filters are very popular in the removal of impulse noise in digital images. The boundary discriminative noise detection (BDND) is shown to supersede existing algorithms especially under high noise densities. However, careful investigation of the filtering step in this algorithm reveals some problems that may affect the sharpness of the output image. In this paper, we propose two simple, yet effective modifications, to the filtering step in the BDND algorithm. The modifications loosen the condition on expanding the filtering window and incorporate the spatial information of pixels in the filtering operations. Experimental evaluation proved the effectiveness of the proposed modifications in producing sharper images than the BDND algorithm.