Image restoration method and system based on conditional generative adversarial network

Abstract

The invention provides an image restoration method and system based on a conditional generative adversarial network. The image restoration method comprises the steps: firstly, training a conditional generative adversarial network CGAN on a CelebA data set, then inputting a to-be-restored image into the CGAN to generate a series of fake images similar to the to-be-restored image, and defining a loss function composed of context loss and perception loss, wherein the context loss ensures the similarity between the image to be restored and the restored image content, and the perception loss guarantees to visually output a complete and vivid image; and finally, mapping the to-be-repaired image to a small potential space by utilizing a back propagation algorithm of the loss function, and inputting the mapped vector into the CGAN to generate an optimal forged image of the to-be-repaired image. The image restoration method has the advantages that the image restoration method based on the conditional generative adversarial network is provided to solve the problems of unstable training and collapse of the generative adversarial network fundamentally and thoroughly, and obtain an optimal forged image in combination with context loss and perception loss functions to complete image missing area restoration.

Description

technical field [0001] The invention relates to the field of computer vision image restoration, in particular to an image restoration method and system based on a conditional generation confrontation network. Background technique [0002] Image restoration originated from people’s restoration of artworks during the Renaissance. Reconstruction of beautiful artworks inherits national culture for people to appreciate and learn. Using the neighborhood information of the missing area of ​​​​the image, the missing area is restored according to certain restoration rules, so that the observer There is no visual perception that the image has ever been damaged or has been repaired. Since the 1980s, with the rapid development of computer artificial intelligence and digital media technology, the restoration of works of art has changed from traditional pure manual restoration to computer automatic detection of damaged areas and completion of restoration. Digital image restoration techno...

AI Technical Summary

Problems solved by technology

However, the existing image restoration algorithms based on deep learning are all based on supervised learning, which brings many restrictive factors to image restoration.

In the field of unsupervised learning, the generative adversarial network (GAN) proposed by Goodfellow in 2014 has made groundbreaking progress. In the process of image repair, the generative adversarial network can be better than the encoder-decoder. Fitting data, the fitting speed is faster and the generated samples are sharper, but this method also has many disadvantages, such as unstable data training, uncontrollable model freedom, training collapse, etc.

Images