Large-area damage image restoration algorithm based on generative adversarial network

Abstract

Given that the traditional image restoration algorithm cannot generate high-quality false images and the restoration accuracy for the large-area damaged images is low, this study proposed the restoration algorithm of large-area damaged images based on the generative adversarial network. First of all, this study extracted multi-scale edge detailed information in image damage area through building the smooth function. Secondly, this study built the generative adversarial network model by using the multi-scale edge detailed information as the original information of large-area damaged images and then trained the model to generate the best false images through the continuous game between the generator and discriminator. Finally, this study obtained the existing information of the false images by combining the contextual information and the perceptual information, calculated the priority of information and restored the information according to the optimal sequencing results and continuously updated the damaged edge information until the large-area damaged image restoration is completed. The results show that the accuracy of extracting the detailed information with the proposed algorithm is high, and peak signal-to-noise ratio of the false images generated by the generative adversarial network model is high, the structural similarity index with the real image is not less than 0.93, the quality of the false images is high, the accuracy rate of priority calculation of the restoration information is high, and the restoration accuracy for large-area damaged images is far ahead of other algorithms.