Abstract: In face recognition,efficient feature extraction method is the key．Canonical correlation analysis (CCA) is a classic feature extraction method,but due to the singularity of the covariance matrices of its two groups of features caused by the small sample high-dimensional feature problem,traditional CCA fails．Moreover because its globally linear property in nature,it can not better portray the local changes in the face image．So there are some two defects of poor prediction accuracy and stability．To improve the prediction accuracy of face recognition model,a novel face recognition method was proposed based on sub-pattern CCA (SpCCA)．With the correlation between global features and local features,the redundant information between the features was eliminated,and the global information and local information were integrated effectively at the same time．Lastly,SpCCA was applied to AR and Yale datasets,and was proved to have significantly better recognition accuracy and higher stability in contrast to the reference model．The result shows that SpCCA can avoid the small sample and nonlinear problems with the assistance of sub-pattern.

Key words: Face recognition,Canonical correlation analysis,Sub-pattern,Fusional features