A Novel Dual-Modal Biometric Recognition Method Based on Weighted Joint Group Sparse Representation Classification

Chunxin Fang, Hui Ma, Yu Li

Abstract: Multi-modal biometric recognition technology is an effective method to improve the accuracy and reliability of identity recognition. However, there are some problems (such as feature space incompatibility) with the fusion between different modal biometric traits. To address the above problem, we propose a dual-modal biometric recognition method based on weighted joint group sparse representation classification (WJGSRC). The proposed method fuses the Pyramid Histogram of Oriented Gradients (PHOG) feature and Local Phase Quantization (LPQ) feature for each modality by the Canonical Correlation Analysis (CCA) at first. Then, the dictionary matrix is optimized by the sum of weighted scores between different modalities. Finally, the group sparse and weight constraints are constructed respectively to further improve the final recognition accuracy. The experimental results on two dual-modal databases show that the proposed method can effectively improve the performance of identity recognition.