改进的用于人脸对准的多尺度ASM方法

Improved Multi-resolution Active Shape Model for Face Alignment

针对传统的ActiveShapeModel(ASM)方法的缺陷与不足,提出了几点改进的方法。传统的ASM方法很难在每个特征点上都得到比较准确的结果,并且,传统的ASM方法的匹配结果受初值的影响非常大。根据人脸的结构特征,把ASM的匹配过程和人脸的部件定位信息以及人脸的边缘信息等相结合,有效地改善了ASM方法的性能。实验结果表明,改进的ASM方法在准确性和鲁棒性上有较大提高。

Several improvements were proposed on the conventional Active Shape Model (ASM) for face alignment. It is hard for the conventional ASM to get accurate result on each landmark, what's more, the performance depends heavily on the initial positions of the landmarks. According to the structure of face, we introduced the edge information and part information of face to the matching process of ASM which improved the performance of ASM. Experiments on a database containing 200 labeled face images show that the improved ASM is more accurate and robust than the conventional one.