The Research And Implementation On Facial Feature Extraction Based On Active Shape Model

To make the identification on suspects through their face characteristics has always been difficult to investigators. This not only causes timing delay of detection, but also increases the workload of investigators.The improvement of urban surveillance coverage makes the access to the suspect’s facial features and even a clear face image possible. At the same time, photos for second generation ID card with uniform standards have been registered. Through the creation of library for face images with feature extraction, we can use the computer to make the identification automatically based on suspects’facial features.Based on the study on facial feature extraction techniques, the main work of this paper are:(1) Research and implementation on Active Shape Model (ASM).Active shape model is method for face feature extraction based on statistical knowledge, using two sub-models for point local search and global shape constraint, which achieves well extraction results.(2) Improvement of Active Shape Model. The iteration pattern of two sub-models in ASM causes offset to feature points, which reduces extraction accuracy. For the reasons causing this problem, we propose refined algorithm, using profile model to adjust point position after shape model. As a result, improved extraction accuracy is achieved.(3) ID photos feature extraction optimization. We propose a background subtraction algorithm to obtain accurate face shape. In addition, nostril shape is extracted to enrich face classification criteria.