When using a regularized approach for image restoration there is always a compromise between image sharpness and noise suppression. Therefore, the main problem is to remove as much noise as possible while preserving sharpness in the restoration. To this cause we introduce a spatially regularized neural approach that makes use of local image statistics to apply varying regularization to different areas of the image. This is achieved with an efficient parallel implementation of the Hopfield neural network. The proposed approach exhibits an improvement in restoration quality and execution time over the existing approaches. This is illustrated on simulations on benchmark images.
展开▼
