Training and reconstruction method of super-resolution reconstruction model of face image

Abstract

The invention relates to the technical field of image processing, and concretely relates to a training and reconstruction method of a super-resolution reconstruction model of a face image. The training method comprises the following steps: obtaining a first sample image and a second sample image corresponding to the first sample image, wherein the resolution of the first sample image is higher than that of the second sample image; inputting the second sample image into a super-resolution reconstruction model to obtain a first reconstructed image; calculating prior information of the first reconstructed image and the first sample image, and determining face prior information loss; and updating parameters of the super-resolution reconstruction model according to the face prior information loss so as to determine a target super-resolution reconstruction model. The model is trained by combining the prior information of the image, a more real face can be obtained, the five sense organs areclearer in contour, the expression is very natural, and the face super-resolution effect is effectively improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a training and reconstruction method of a super-resolution reconstruction model of a face image. Background technique [0002] In some crowded places, such as playgrounds, large shopping malls, etc., illegal incidents are prone to occur, so it is often necessary to find some people who may have special identities from surveillance images. However, due to objective reasons such as unstable ambient light, diverse postures and expressions of pedestrians, a relatively long distance from the camera, and limitations in the imaging resolution of the camera, the pedestrians in the image cannot be accurately identified. As the most sensitive area of ​​human identity features, the face has become the most critical part of specific identity recognition. Therefore, how to reconstruct a face with low resolution, low definition, noise and light pollution into a high-resolution, high-d...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the embodiment of the present invention provides a training and reconstruction method of a super-resolution reconstruction model of a human face image to solve the problem that the super-resolution reconstruction method in a general scene is directly applied to a human face image and the reconstruction efficiency is very low. The problem

Images