Facial feature point extraction method based on combination of shape extraction and pattern matching

In this paper, we propose a method for fast and accurate extraction of feature points such as pupils, nostrils, mouth edges, and the like from dynamic images with the purpose of face recognition. Accuracy of face extraction with these feature points used as criteria greatly affects the capabilities of face recognition methods based on pattern matching. This processing must be performed rapidly, con-sidering the large number of recognition trials required for dynamic images and the requirements of real-time registra-tion. In various extraction methods proposed in the past, steady extraction was difficult due to influences such as individual differences, expression variations, face direction or illumination variations, and so forth. These methods are far from satisfactory in terms of extraction accuracy and processing speed. The proposed method achieves high po-sition accuracy at a low computing cost by combining shape extraction with pattern matching. In particular, it uses a separability filter to extract feature point candidates for pupils, nostrils, mouth edges, and the like. Next, it uses pattern matching based on the subspace method to select the correct feature points from the candidates. Results of testing facial images under various conditions using an evaluation system demonstrated that for 1700 static images the feature point extraction rate was 99%, and in the case of dynamic images the extraction rate for 9880 frames was 98 % at a speed of 10 trials/s, without using hardware