Face image recovery method and device based on cyclic neural network

Abstract

The invention discloses a face image recovery method and device based on a cyclic neural network, and the method comprises the steps: obtaining a low-resolution image, carrying out the pedestrian facedetection of the low-resolution image, and obtaining a to-be-restored face image; inputting the to-be-restored face image into a pre-built and trained neural network model, and extracting a feature map of the to-be-restored face image according to a preset pixel size; carrying out the nonlinear mapping of the extracted feature map, and obtaining an optimized feature map; carrying out the high-dimension reconstruction of the to-be-restored face image according to the optimized feature map, and outputting a corresponding high-resolution face image. According to the invention, the bidirectionalcyclic neural network technology and the generative adversarial network technology are combined for constructing and training a neural network model with the optimal recovery performance, so the method can achieve the effective recovery of a high-resolution face image from the low-resolution face image.

Description

Technical field [0001] The present invention relates to the technical field of face recognition, in particular to a method and device for face image restoration based on cyclic neural network. Background technique [0002] At present, there are many mainstream face image restoration methods, which can be divided into three categories according to the different ways of obtaining the target image: 1. The traditional algorithm based on interpolation, the principle of which is mainly by performing multiple images in the same scene Contrast, and obtain the corresponding sub-pixel information from these images, and then obtain the sample information of the sampling points through the calculation of these sub-pixels; 2. Based on the reconstruction algorithm: by studying the high-resolution details of the image at low resolution The form of expression, establish the corresponding relationship between the two, and use a certain model to describe this mapping relationship; 3. Based on trad...

AI Technical Summary

Problems solved by technology

[0003] There are some defects and deficiencies in the above three existing face image restoration methods: the image restored by the method based on interpolation does not increase the image information, and the restoration ratio is low; using the reconstruction-based algorithm, as the restoration ratio increases, it needs The number of input image samples provided increases sharply, until it reaches a certain upper limit, no matter how many input image samples are added, the reconstruction effect cannot be improved; the face recovered based on traditional machine learning methods is not clear enough, only the outline of the face, No clear facial features

Images