Spatio and Efficient l1-l1 minimization based Impulse Noise Removal in Gray Images Using Dictionary Learning

摘要

In natural images spatially adjacent image pixels have similar pixel values and many patches of image pixels have similar values. This similarity exploited for reducing the computation time required for de-noising and l1-l1 minimization was modified for efficient implementation. Using impulse noise detector, noisy pixels were separated and from noise free pixels DC values of image batches were calculated. This accurate DC value calculation improves the quality of the de-noised image and preserves the details. Once noise is removed using efficient 11-11 minimization, de-noised pixels will replace noisy pixel in the corrupted image. The proposed algorithm gives superior peak signal-to-noise ratio (PSNR) and structural similarity (SSEM) indices compared with the other state-of-the-art algorithms for grey images.