中文摘要：

活动形状模型（ASM）是一种较为流行的用于目标对象定位的统计学形状模型。然而，在搜索过程中，ASM易受初始化位置的影响而陷入局部极值；在局部灰度模型匹配时，对光照条件也较为敏感。针对正面人脸彩色图像的识别问题，提出了一种基于肤色的人脸检测方法、基于人脸器官几何分布特征的人眼检测方法与活动形状模型方法相结合的算法，来降低ASM算法对初值与光照条件的敏感程度，避免局部极值问题的产生。对于搜索结果的评估问题，本研究针对保持了目标特异性的形状，提出了以质心位置、形状大小和方位角作为相似形状的评价准则。采用40幅黄种人正面端正人脸图像进行测试，实验结果表明，本改进算法的搜索结果与手工标定的识别结果相比，在位置、大小和方位上的误差都较小，识别准确率较高。

英文摘要：

Active Shape Models （ASM） algorithm is a popular statistical shape model method for objects localization. However, the performance of ASM depends heavily on the initial model instance. It is also sensitive to image illumination during the matching of local grey-level model of each landmark. In this paper, we proposed a combined algorithm of ASM, skin color detection and facial features＇ distribution based on human eye detection. With this modified ASM, the influence of initial instance and image illumination could be reduced. Since the searching results by ASM can maintain the specificity to the class of object it intends to represent, the centroid location, shape size and orientation angle were used for evaluating the proposed algorithm. Forty frontal face images of yellow race were used in our experiments, and results showed that the shape location, size and orientation errors of the improved ASM searching results were smaller comparing with the manually annotated face shapes.