Face super-resolution reconstruction method based on extremely deep convolutional neural network

Abstract

The invention provides a face super-resolution reconstruction method based on an extremely deep convolutional neural network, which comprises the following steps of 1, performing down sampling and processing on a high-resolution face image by different multiples to obtain a low-resolution face image training set; 2, performing down sampling and processing on another high-resolution face image by different multiples to obtain a low-resolution face image test set; 3, putting the training set obtained in step 1 and the test set obtained in the step 2 in the extremely deep convolutional neural network for training, learning the mapping of a residual image, and obtaining a corresponding convolutional network model; and 4, inputting a low-resolution face image that needs to be reconstructed intothe convolutional network model obtained in the step 3 to obtain a reconstructed high-resolution face image. The face super-resolution reconstruction method based on the extremely deep convolutionalneural network can better deal with the super-resolution reconstruction problem of multi-scale amplification factors.

Description

technical field [0001] The invention belongs to the technical field of image information processing, and in particular relates to a face super-resolution reconstruction method based on an extremely deep convolutional neural network. Background technique [0002] Image super-resolution reconstruction is a classic problem in the field of computer vision, which aims to infer high-resolution (HR) images with key information from a given low-resolution (LR) image, among which, face super-resolution reconstruction is One of the important branches. Face super-resolution reconstruction has a wide application background in many fields such as identity authentication and intelligent monitoring. The current super-resolution reconstruction methods are mainly divided into two categories: (1) regard super-resolution as an ill-posed problem in image processing, which can be solved by introducing prior information; (2) use machine learning methods to obtain low The mapping relationship be...

AI Technical Summary

Problems solved by technology

[0005] Although CNN has been successfully applied to face super-resolution reconstruction and achieved good results, it also has limitations in the following aspects: first, CNN relies heavily on the context of small image regions; second , as the number of layers increases, the receptive field becomes larger and the details of the reconstructed image are easily lost; third, the training convergence speed is too slow

Images