Detection and removal of random-valued impulse noise from images using sparse representations

This paper proposes a novel iterative method for the removal of random-valued impulse noise from the images using sparse representations. Each iteration has three stages. In the first stage, the positions of the possible noise pixels are detected using a sparse representation of the pixels in a window. In the next stage, the pixels that are detected as noisy pixels are treated as missing pixels and are filled using image inpainting through sparse representations. In the third stage, the pixels detected as noise pixels in the first stage are tested based on the inpainted value to determine the correctness of the noise detection at the first stage. In the subsequent iterations, the output of the previous iteration is considered to be the input for the detection and removal of the impulse noise.