Multi-pose Face Recognition Method Based on Low-rank Decomposition and Sparse Residual Contrast

Abstract

The invention discloses a multi-pose face recognition method based on low-rank features and sparse representation comparison classification. The present invention first uses the dual low-rank decomposition method to perform dimension reduction decomposition and optimization on the input face picture, and obtains the first-type low-rank feature that removes the attitude structure; secondly, the structured irrelevant low-rank decomposition, through the augmented Lagrangian The multiplier method ALM performs alternate iterative solutions to obtain the second-type low-rank features; finally, based on the residual comparison classification of sparse representation: if the classification results of the two features are the same, the classification labels remain unchanged; if the classification labels are not the same, construct The residual rate comparison model compares the ratio of the difference between the second minimum residual and the minimum residual to the minimum residual after two features are sparsely represented. The classification result with higher residual rate in the two features is selected as the final classification category. The low-rank decomposition method and the sparse representation residual comparison model used in the present invention can effectively remove the interference caused by the gesture structure to the recognition effect.

Description

technical field [0001] The invention belongs to the technical field of computer image processing, and relates to a face recognition method based on low-rank decomposition and sparse representation residual comparison under various postures. Background technique [0002] Multi-pose face data is widely used in today's society, and data perception and data acquisition based on different angles can help researchers better represent data. Therefore, face recognition research for multi-pose face data has also become an important development direction in the field of biometric recognition and pattern recognition. [0003] Considering that there are large differences between different face poses, there are usually two structures in a sample, namely the face category structure and the pose category structure. These two class structures are entangled in raw image data. Therefore, how to overcome the interference caused by multi-pose face data on the recognition accuracy and ensure a...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is mainly aimed at the shortcomings of low face recognition accuracy and poor robustness in a multi-pose environment, and proposes a face recognition method with high recognition rate, high robustness and high efficiency

Images