Partial Face Recognition Based on Template Matching

A partial face recognition strategy based on template matching is proposed in this paper. In real-world face recognition applications such as surveillance or forensics images only a small part of face image is available. In the proposed approach in this paper, instead of comparing the small sub-image to the complete face images in the database, a template matching technique finds face parts on the gallery samples in the database with the best match to the partial image. The feature extraction and classification techniques are applied on the small sub images of the probe and gallery sets to find the identity of the partial probe face. AR, LFW and FERET databases are employed to evaluate the performance of the proposed approach. Normalized sum of squared difference (NSSD) outperforms zero mean normalized cross correlation (ZNCC) in template matching. Based on the experimental results, partial eyes image leads to the best recognition accuracy. Also, reducing the size of sub-image to less than 6.25% of the complete image size, decreases the identification accuracy drastically.