An efficient and computationally effective random valued impulse noise detector

The paper proposes an efficient and computationally effective method for detecting random valued impulse noise of images, corrupted with high density. The method works in two stages of operation. The first stage of operation identifies the possible candidates for noisy pixels through the decision boundaries along with some prior knowledge. Second stage calculates the difference of maximum and minimum absolute values of the four convolutions, using four one-dimensional Laplacian operators to judge the exact noisy pixels. As a result, the cumulative effect of these operations enhances the noise detection capability of the method with minimum error, which supports the next step of filtering operation in preserving fine lines and edge details. Superiority of the proposed method to various existing methods in terms of miss detection rate and false alarm rate, is experimentally verified with ten different real-time images corrupted with random valued impulse noise of high densities. In addition, the proposed method performs detection process efficiently with less computational complexities, and thus claims its potentiality for online applications.