Human face recognition method based on bi-directionally and two-dimensionally iterative and non-relevant discriminant analysis

Abstract

The invention discloses a human face recognition method based on the bi-directionally and two-dimensionally iterative and non-relevant discriminant analysis. Firstly, the image sample of a human face is acquired. After that, within-class scatter matrixes, inter-class scatter matrixes and overall scatter matrixes in the horizontal and vertical directions are respectively calculated. By utilizing the non-relevant discriminant transformation method, a first optimal differential vector and an optimal discriminant vector set in each direction are respectively calculated. Afterwards, the projection and dimensionality reduction of the image is conduced by the two optimal discriminant vector sets at the same time. Finally, the classified calculation is conducted by a nearest neighbor classifier to figure out the recognition rate of the image. According to the technical scheme of the invention, the non-relevant discriminant transformation on two-dimensional images is conducted based on the discrimination information in two directions. In this way, the dimensionality reduction and the discrimination information extraction of two-dimensional images both in the horizontal direction and in the vertical direction can be realized at the same time. Therefore, the correct facial recognition is realized. Both the processing time and the storage space are saved, and the recognition rate is ensured to be high.

Description

technical field [0001] The invention relates to the field of face recognition technology and pattern recognition, in particular to a face recognition method of bidirectional two-dimensional iterative non-correlation discriminant analysis. Background technique [0002] Face recognition technology is a kind of biometric recognition technology based on human facial feature information. It has a wide range of fields in mathematics, computer, automation, visualization, electronics, virtual reality, pattern recognition and image processing. Research. It has been widely used in government, finance, military, justice, public security, border inspection, electric power, aerospace, education, factories, medical care and many enterprises and institutions. It is non-mandatory (no need to cooperate with face collection equipment), Non-contact (no direct contact with face acquisition equipment), concurrency (can operate on multiple faces at the same time), etc. are superior to other biom...

AI Technical Summary

Problems solved by technology

[0004] When using the Fisher (Fisher, name) criterion function to solve the optimal discriminant vector set, a condition that must be satisfied is that the intra-class scatter matrix is ​​non-singular. To ensure the establishment of this condition, it is necessary to ensure that the number of samples participating in the training is not less than In fact, when using one-dimensional linear discriminant analysis for feature extraction, the two-dimensional face image matrix must first be stretched into a one-dimensional vector by row or column, and then calculated, which not only destroys the original The structure of the two-dimensional face image, and the dimension of the sample is too high (up to tens of thousands of dimensions), so that the number of samples is much smaller than the dimension of the sample when performing face recognition. Therefore, it is easy to cause the intra-class divergence matrix Singularity, the so-called "small sample size" problem

In order to avoid the "small sample" problem, feature extraction can be performed directly on the two-dimensional face image matrix. The specific methods include two-dimensional linear discriminant analysis (2DLDA), two-dimensional principal component analysis (2DPCA) and two-dimensional non-correlated discriminant transformation ( 2DUDT), etc., but these three methods require more coefficients, larger dimensions, longer calculation time, larger storage space, and generally only extract discriminant information in one direction, and only perform dimensionality reduction in one direction. However, the discriminative information in the other direction is ignored, resulting in inaccurate recognition.

Images