| Summary: | 碩士 === 國立成功大學 === 電腦與通信工程研究所 === 98 === The 3-dimensional (3D) face model based on two-dimensional images has been applied to many situations, such as facial recognition, facial surgery simulation, and 3D animation. Although a great deal of literature has addressed the construction of face model, it is restricted to define facial feature points manually. The purpose of this paper is to automate the process of 3D model construction.Active Appearance Model (AAM) automatically extracts facial features is adopted. In addition, a simple method of estimating initial position of AAM increases the accuracy of extraction is introduced. Based on feature points position obtained from AAM, 3D Morphable Model is aligned to face on the image. Accordingly, maximum a posteriori estimator (MAP) will sort out the best model parameters such that the appearance obtained from 3D face model is resemble to the test image. For automatic extracting face features from frontal face image, this paper has better accuracy than Smallest Univalue Segment Assimilating Nucleus (SUSAN) in different light sources environments and race of human.
|
|---|
