Edge-preserved neural network model for image restoration

This paper presents a combined approach for image restoration with edge-pre serving regularization, subband coding, and artificial neural network. The edge information is detected from the source image as a priori knowledge to recover the details and reduce the ringing artifact of the subband coded i mage. The multilayer perceptron model is employed to implement the restorat ion of images. The main merit of the presented approach is that the neural network model is massively parallel with stronger robustness for transmissi on noise and parameter or structure perturbation, and it can be realized by very large scale integrated technologies for realtime applications. To eva luate the performance of the proposed approach, a comparative study with th e set partitioning in hierarchical tree (SPIHT) has been made by using a se t of gray-scale digital images. The experiment has shown that the proposed approach could result in considerably better performances compared with SPI HT on both objective and subjective quality for lower compression ratio sub band coded image. (C) 2001 SPIE and IS&T.