Weight assignment for adaptive image restoration by neural networks

This paper presents a scheme for adaptively training the weights, in terms of varying the regularization parameter, in a neural network for the restor ation of digital images. The flexibility of neural-network-based image rest oration algorithms easily allow the variation of restoration parameters suc h as blur statistics and regularization value spatially and temporally with in the image. This paper focuses on spatial variation of the regularization parameter. We first show that the previously proposed neural-network metho d based on gradient descent can only find suboptimal solutions, and then in troduce a regional processing approach based on local statistics. A method is presented to vary the regularization parameter spatially. This method is applied to a number of images degraded by various levels of noise, and the results are examined, The method is also applied to an image degraded by s patially variant blur, In all cases, the proposed mettled provides visually satisfactory results in an efficient way.