Abstract: In order to improve the clarity of the blind restoration of the conformance fuzzy image, a prior fuzzy blind restoration algorithm based on the prior optimization is proposed for the study of the prior constraint problem of the full variational model involved in the restoration process. Firstly, the local weighted total variation model based on half Gauss gradient operator is used to extract the significant edge of the blurred image. The noise and texture interference are removed, and the ability to maintain the favorable information is improved. Then a multi-scale mixed characteristic prior estimation of blur kernel is proposed to enhance the accuracy of blur kernel estimation. Finally, clear restored images are obtained by non-blind deconvolution. The experimental results show that compared with other algorithms, the proposed algorithm improves the average peak signal to noise ratio of the reconstructed image by about 1.7%, and the average structure similarity index increases by about 19.1%. In view of the real blur image, the artifact of restored image is less, the edge texture details are more clear and natural, and the overall visual effect is better.

Key words: half Gauss gradient operator, multi-scale mixing characteristics, depth convolution neural network prior, consistency blur, image blind restoration