A face recognition method and device based on tensor description

Abstract

The invention discloses a face recognition method and device based on tensor description. Firstly, similarity learning is performed on labeled image samples and unlabeled samples to be classified, and similar neighbor graphs and normalized weights are constructed for use in Characterize sample similarity, and then manually initialize a class label matrix. In order to effectively realize the direct induction of out-of-sample face images, the present invention integrates a tensor-based regularization term that can directly induct out-of-sample images into the existing label propagation model, and finally use the parameters to weigh the similarity measure, the initial category label and the influence of the regularization item based on the matrix pattern on the face recognition, complete the system modeling, take the maximum value of the similarity probability in the system output, and use it for the face image category identification, to obtain the most accurate system identification results. By introducing the idea of ​​tensor description, the topological structure between image pixels can be effectively maintained in the induction process of out-of-sample face images, and the system has good scalability.

Description

technical field [0001] The invention relates to the technical fields of machine learning, pattern recognition and data mining, in particular to a face recognition method and device based on tensor description. Background technique [0002] In the past few decades, face recognition technology has become a very popular research topic in computer vision. In recent years, face recognition technology has developed rapidly, and many devices have been put into use and have produced huge social and economic benefits. Face recognition methods mainly include geometric feature recognition, eigenface-based recognition, neural network recognition and support vector machine recognition, etc., while label propagation technology is still rarely used in the field of face recognition. [0003] Label propagation is a semi-supervised learning method based on similarity graph construction. Among many semi-supervised learning methods, label propagation method has attracted widespread attention b...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the present invention provides a face recognition method and device based on tensor description to solve the problem that the vector space model in the prior art will destroy the topological structure between image pixels, resulting in reduced recognition accuracy and model training and prediction processes. Highly complex technical issues

Images