Face image super-resolution reconstruction recognition method based on multi-set canonical correlation analysis

Abstract

The invention discloses a face image super-resolution reconstruction recognition method based on multi-set canonical correlation analysis. The method comprises the following steps of: 1, simultaneously learning mapping of various specific resolutions for various homologous low-resolution human face views by using multi-set canonical correlation analysis, dividing a multi-view human face image intooverlapped image blocks, and extracting potential coherent features of principal components of the image blocks by using the multi-set canonical correlation analysis; 2, reconstructing a global faceimage by using a neighborhood reconstruction strategy, and supplementing details to the global face image through residual compensation; and 3, the finally output super-resolution reconstruction result is the sum of the global face image and the residual face image. The method has a certain theoretical basis, test results in different databases show that the method has high robustness and certainmarket implementation feasibility, the multi-view super-resolution problem which cannot be processed by most of existing algorithms at present is solved, and the method has high innovativeness and practicability.

Description

technical field [0001] The invention relates to the field of super-resolution reconstruction and recognition, in particular to a face image super-resolution reconstruction and recognition method based on multi-set canonical correlation analysis Background technique [0002] Multivariate analysis methods are often used in super-resolution reconstruction for feature extraction, among which principal component analysis (Principal Component Analysis, PCA) and canonical correlation analysis (Canonical Correlation Analysis, CCA) are more popular. The feature extraction step is often used to reduce the dimensionality of the data and reduce noise. PCA extracts useful information of faces and filters noise by retaining appropriate dimensions. Wang et al. proposed a framework for generating high-resolution faces by obtaining linear combination coefficients of images through PCA. Huang et al. proposed a super-resolution method using CCA to extract the relationship between high and lo...

AI Technical Summary

Problems solved by technology

[0005] However, in real life, due to distance or device limitations, people usually need to face the same face with multiple different resolution views. For multiple different low resolution views, most existing methods cannot be effective. To deal with this kind of problem, they cannot simultaneously learn the mapping relationship that maps multiple low-resolution views to high-resolution views at the same time

To date, jointly learning the relationship between multiple views has not received much attention

Images