Shielded face restoration method based on geometric perception priori guidance

Abstract

The invention discloses an occluded face restoration method based on geometric perception prior guidance. The method comprises the following steps: S1, establishing a face semantic analysis module; s2, inputting the shielded face image into a face semantic analysis module to obtain a face semantic analysis graph; s3, splicing the face semantic analysis graph, the sheltered face image and the random sheltered Mask into a five-channel picture, and taking the five-channel picture as the input of the repairing network generator in the step S4; s4, constructing a repair network generator to obtain a final repair image; s5, inputting the restored image into a global discriminator and a local discriminator; and step S6, carrying out an experiment on the public data set by using the designed method, and testing three indexes, namely, a Peak Signal-to-Noise Ratio, a Structural Signal Index and a Frechet Inception Distance. The method provided by the invention has the following beneficial effects: the method provided by the invention can be used for carrying out an experiment on the public data set; and the method provided by the invention can be used for testing the three indexes, namely the Peak Signal-to-Noise Ratio, the Structural Signal Index and the Frechet Inception Distance.

Description

technical field [0001] The invention relates to image processing technology, in particular to an occluded face restoration method based on geometric perception prior guidance. Background technique [0002] Image inpainting technology is to use the information of the non-missing areas in the damaged image to repair and fill the contaminated or missing areas in the image according to certain repair rules. This is the advantage of deep neural networks. At present, the most representative generative model is Generative Adversarial Networks (GAN). [0003] Nowadays, with the development of technology and social needs, face recognition technology has been further developed. However, when a face image is acquired, it is not only affected by the environment, but also the face is occluded. These factors will affect the accuracy of face recognition. With the development of the pneumonia epidemic sweeping the world, people's public health awareness has been greatly improved, and the...

AI Technical Summary

Problems solved by technology

Existing methods propose a network including LSTM decoder and dual-channel LSTM decoder for occlusion detection and repair, and introduce classification loss to minimize identity information loss

Although some results have been achieved, the following problems remain unresolved: the calculation of the loss function is based on the feature map of the entire face, but the occlusion only occupies a part of the face, so it lacks pertinence; the calculation of the loss function generally uses L1 Or the L2 norm, which will make the repaired face tend to the average face of the data set, which is not conducive to subsequent recognition; repair is an estimation process, unreliable, and repair is inevitable, but there is currently no evaluation of the reliability of the repair method, resulting in a lack of theoretical support for repair-based applications

Images