Development and testing of a LBP-SVM based teeth visibility recognizer

Human face recognition receives more and more attention for its important role in a wide range of areas such as image searching, video surveillance, and human-computer interaction. This paper focuses on developing and testing a working attribute recognizer for one specific facial characteristic - teeth visibility. Three major steps - image preprocessing, feature extraction and classifier training are involved in the development process. For comparison, both the Local Binary Patterns (LBP) features and features derived from normalized cross-correlation (NCC) template matching are extracted and used to train a SVM classifier. After development, the attribute recognizer is tested with various parameter settings and under several different conditions in this report. Experimental results show that the LBP-SVM-based teeth visibility recognizer has a high accuracy and performs differently under different parameter settings such as the block size, sampling radius, and sampling density and is robust to pose, illumination, and small expression changes. Besides, the LBP-based teeth visibility recognizer is generally superior to that based on normalized cross-correlation template matching. The reasons are also explored in the experiment part.