Face super-resolution method based on multi-scale attention residual error and equivariant mapping

Abstract

The invention discloses a human face super-resolution method based on multi-scale attention residual errors and isovariant mapping. The method comprises the following steps: firstly, extracting shallow layer features of a low-resolution human face side image through a convolutional layer; then, inputting the shallow layer features into a feature extraction subnet, and obtaining deep layer features through n multi-scale attention residual modules; further inputting the obtained deep features into a residual isovariant mapping module, combining the deep features with residual features fused with yaw coefficients in a deep representation feature space, and transforming feature vectors of the side face into a feature vector space the same as that of the front face; and finally, a high-resolution face front image is obtained through a reconstruction module. The method is suitable for face recognition, does not excessively depend on a large number of front face and side face data pairs any more, and can reconstruct a front face image with high definition and more real texture for an input low-resolution side face image.

Description

technical field [0001] The invention belongs to the field of image processing and computer vision, and relates to a super-resolution reconstruction method of a face image, in particular to a face super-resolution method based on multi-scale attention residuals and equivariant mapping. Background technique [0002] Thanks to the advent of deep learning, face recognition has achieved extraordinary success. However, many existing face recognition models still perform relatively poorly when processing images of profile faces compared to images of frontal faces. And in practical applications, the resolution of the obtained face images is relatively low, even if converted into a frontal face, the effect of the image is relatively poor. [0003] In order to solve the above problems, the present invention proposes a face super-resolution method based on multi-scale attention residuals and equivariant mapping, in order to realize super-resolution reconstruction of low-resolution sid...

AI Technical Summary

Problems solved by technology

[0004] The present invention is used to solve the problem that the current image super-resolution reconstruction method can only simply reconstruct the image super-resolution into a high-resolution image, but cannot realize the angle conversion of the image from the front face to the side face

Images