Robust Part-based Face Matching with Multiple Templates

Part-based matching can handle significant variations in facial expression and partial occlusion. However, under large pose variations, it is sometimes difficult to find corresponding local parts in a pair of images. In this paper, we propose a novel part-based face matching scheme that relies on multiple face templates. These templates serve as bridges to help finding corresponding parts between test images with different face poses, thus making the overall algorithm more robust compared with traditional schemes. Experimental results on Multi-PIE public benchmark data set show that our method is very effective in improving face matching performance with significant pose variations.