| Summary: | 碩士 === 國立東華大學 === 資訊工程學系 === 100 === Facial component extraction (FCE) is a key problem in the facial information processing and analyzing (FIPA) because many facial applications require FCE as a preprocessing step. For extracting facial components, the active shape model (ASM) is an option commonly adopted by many researchers. As reported in many studies, the efficacy of ASM significantly depends on the initializing positions of facial landmarks and the finally searched facial landmarks. Badly initialized facial landmarks would not lead to accurate searching of facial landmarks and then cause inaccurate extraction of the contours of facial components. Focusing on the above key problems, the study presents the improvement over the traditional ASM by proposing the following design: (1) an iterative initialization method to incrementally refine the initial facial landmarks, (2) a facial landmark positioning method based on dynamic time warping algorithm, and (3) a component-based modeling of individual facial components, instead of the whole-face modeling used in traditional ASM. By conducting a number of experiments on trained and un-trained faces, the proposed method indeed achieves better accuracies of facial component extraction than the traditional ASM. Moreover, the results show that the proposed component-based approach is better than the common whole-face approach.
|
|---|
