Research On Human Facial Feature Points Positioning

Face recognition is a hot issue in computer vision and pattern recognition domain. Face recognition has promoted the development of many subjects, such as image processing, pattern recognition, computer vision and so on. Facial feature points positioning is automatically getting its main face outline the key characteristics description in any facial image. Facial feature points positioning is the key step of face recognition and the accuracy of positioning will influence the reliability of the subsequent applications directly. The research of facial feature points positioning is widely used in many fields such as face expression analysis, face recognition and face synthesis, face image coding, etc.Active Shape Model (ASM) which based on Point Distributed Model (PDM) is mainstream algorithm in the field of facial feature points positioning. Practice shows that ASM algorithm has higher accuracy and reliability. The paper discusses the process of ASM algorithm in detail: training for facial image specimen, construction of shape model and searching for target facial feature points. Meanwhile, ASM has its disadvantages: the accuracy of algorithm relies highly on the original position of average shape model in the target facial image, moreover, in the process of searching for target facial feature points, many important texture information is lost easily because local gray level information near the fixed point is adopted.In consideration of Gabor wavelet's analytical ability of time-frequency local texture and above mentioned disadvantages of ASM algorithm, the improvement of ASM algorithm is put forward as follows : (1)Face area is detected in target facial image, thus , the original position of shape model is perfected. (2) With the introduction of Gaussian distribution, the weighting gray level model is constructed. Therefore, the loss of important texture information is avoided. (3)Construction of local Gabor feature model of fixed point is proposed. Based on ASM positioning, the positioning results are updated through local Gabor feature model. Experiment shows that the improved ASM algorithm does improve the accuracy of facial feature points positioning to a certain extent.