Time-Lapse Data Oriented Infrared Face Recognition Method Using Block-PCA

This paper copes with infrared (IR) face recognition on time-lapse data, which result in significantly decline of recognition rate. In order to eliminate the effects of the ambient temperature, psychological and physiological factors on infrared imaging, the block-PCA is proposed for feature extraction. The method calculates the standard deviation of each principal component, which is utilized to determine which principal component is discarded. To further improve the performance, the infrared thermal images are converted into blood perfusion images to get more stable biological features, based on which the block-PCA is performed. Experimental results on time-lapse data show that the proposed approach achieves 30.3% higher in recognition rate than the conventional PCA.