Successive difference detection based adaptive iterative median filter for image restoration

Abstract

The paper proposes a universal noise removing filter capable of removing a mix of salt and pepper, random and Gaussian noise. In addition, the filter can also remove degradations like scratches, streaks, grids etc. that may corrupt images in real time. The algorithm exploits the fact that noise is a violation of spatial coherence of image intensities. In the detection phase, the corrupted and uncorrupted pixels are identified by computing the successive difference of the sorted pixels in the detection window. The correction phase uses the adaptive median filter that is iterated through the image until all the noise pixels are restored. The iteration with the window dimension not exceeding 5 x 5 ensures better preservation of image details. For an image that is corrupted with Salt and Pepper noise of density 60%, Random noise of density 20% and Gaussian noise of Standard Deviation 20, the image restored by this filter has a PSNR as high as 22 dB. The best feature of the proposed Successive Difference Detection Based Adaptive Iterative Median Filter (SDD-AIMF) is the graceful degradation in performance as the noise density increases, which is not the case with popular algorithms. The quantitative and qualitative results clearly prove that the proposed algorithm has better image restoration capabilities than many other popular techniques in literature.