G-ASM: Facial Feature Tracking Based On Gabor Wavelet And Actvie Shape Model

Active Shape Model is a powerful statistical tool to locate and track facial feature points by shape, which has been improved for many years. However, due to some defects in local appearance model, ASM can suffer from changes in illumination and facial expression changes.Gabor wavelets have been applied in computer vision for many years, their power and ability has been proved in many works. In this paper, we designed so-called G-ASM, which combines Gabor wavelet and Active Shape Model to locate and track facial points.Compared with traditional ASM, G-ASM bases on Gabor wavelet instead of local appearance model to guess better places for landmarks, which provide accurate guidance for place-guess. Experimental results demonstrate that G-ASM's ability for facial future tracking.The frameworks and algorithm implement of G-ASM is presented in this paper. How accurately a facial point is guessed based on different Gabor Jets search methods is tested. The performance of G-ASM subjects to the choice Gabor wavelet. An experiment to explore how four different Gabor wavelets effect the performance of G-ASM is presented.G-ASM used only 2D shape information. Therefore, this approach is generic and applicable to track other textured objects.