Improved decision median filter for video sequences corrupted by impulse noise

The recent advances in sparse representations of images have achieved outstanding results in terms of denoising and restoration; but removal of real and structured noise in digital video sequences remains a challenging problem. Based on this idea, the problem addressed in this paper proposes to improve the decision median filtering algorithm for denoising of video sequences corrupted with impulse noise. The proposed algorithm processes the extracted frame (from corrupted video sequences) by incorporating robust decisions to selectively operate upon the corrupted pixels. The local statistical parameters (of the spatial kernel) are then used to decide whether to restore the centre pixel with median value or adaptively increment the kernel size. This helps in restoration of structural content with minimal blurring at high noise densities. Experimental results show that the proposed algorithm achieves better performance with minimal computational complexity; yielding higher values of PSNR and SSIM for restored frames.