基于联合稀疏描述的多姿态三维人脸识别

摘要

A multi-pose 3D face recognition method based on joint sparse representation, called Joint Sparse Repre-sentation-based Classification ( JSRC) , is proposed in this paper. Multi-view 3D face test data are jointed for identi-ty recognition by the hypothesis of multi test data joining the same sparse pattern. Consequently, we, using the JS-RC method, construct 3D overcomplete dictionary and sparse representation model, thus completing the joint recon-struction of the sparse representation vector. The most notable advantage of the JSRC method is:utilizing the corre-lation of multi-view face, reducing the error identification risk of the traditional methods which consider only one test face each time, and improving the recognition accuracy. Experimental results on FRGC2�0 database and their analysis show preliminarily that JSRC method has higher performance in multi-pose 3D face recognition as compared with those obtainable respectively with mutual subspace method and sparse representation-based classification meth-od.%提出一种基于联合稀疏描述的多姿态三维人脸识别算法。该算法基于多幅不同姿态的三维人脸测试样本联合完成身份识别，通过假设多幅测试样本共享同一稀疏类型，联合多视图信息，构建三维空间字典和稀疏描述模型，用于对稀疏描述向量进行联合重建。该方法最显著的特点就是利用所有观测视图的相互关系，避免单独对待每一个观测值时所潜在的错误判别风险，从而提高识别准确率。在国际三维人脸数据库FRGC2�0上的实验证明该算法对多姿态三维人脸的识别性能优于相互子空间算法和稀疏表示识别算法。