Semi-supervised face recognition method based on local information and group sparse constraints

Abstract

The invention discloses a semi-supervised face recognition method based on local information and group sparse constraints. The method comprises the following steps that: obtaining a face dataset X which is an element of a set Rd*n, wherein the face dataset X contains n pieces of high-dimension data, d is a data dimension, the face dataset comprises m marked datasets X1 which is an element of a set Rd*m and a corresponding label matrix Y1 which is an element of a set Rm*c, and c is the classification number of face data; on the dataset X, constructing an unsupervised face feature selection model based on local information constraints; on the marked dataset X1, constructing a supervised face feature selection model based on a matrix l2,1 loss function; constructing a face feature selection target function of the group sparse constraint; utilizing an iterative optimization algorithm to solve the target function; and taking the screened face feature as the input of an SVM (Support Vector Machine), carrying out training to obtain an SVM classifier, and finishing face recognition. By use of the method, the selection and identification accuracy of face features can be effectively improved, and meanwhile, the interference of noise in the dataset can be effectively inhibited.

Description

technical field [0001] The invention belongs to the technical field of machine learning, in particular to a semi-supervised face recognition method based on local information and group sparse constraints. Background technique [0002] Face recognition is an important biometric technology for identity recognition. It has been widely used in various fields, such as intelligent access control, human-computer interaction, and authority management. In recent years, with the rapid development of multimedia technology and network technology, face data has gradually presented high-dimensional features. The processing of these high-dimensional face data not only consumes a lot of time and space resources, but also contains a lot of redundant data. The remaining information brings challenges to face recognition methods based on this type of data. [0003] Although the existing face recognition methods have feature screening of input data, such as the selection of representative featu...

AI Technical Summary

Problems solved by technology

In recent years, with the rapid development of multimedia technology and network technology, face data has gradually presented high-dimensional features. The processing of these high-dimensional face data not only consumes a lot of time and space resources, but also contains a lot of redundant data. The remaining information brings challenges to face recognition methods based on this type of data

[0003] Although the existing face recognition methods have feature screening of input data, such as the selection of representative features by using methods such as principal member analysis (PCA) and FISHER, there are mainly two problems in these methods: (1) These methods only consider the relationship between the same data features, but lack the analysis of the correlation between different data features, making it difficult to screen out the most representative features; (2) These methods are easily affected by external noise, Especially when there are a lot of unlabeled data in the data set, if the interference of noise cannot be effectively suppressed, the effect of face recognition will be seriously affected

[0004] Based on the above analysis, the accuracy of the existing face recognition methods is not high and needs to be improved.

Images