基于ASM的人脸面部特征定位算法的改进

摘要

This paper makes several improvements of traditional active shape model for face orientation.It is hard for the conventional ASM to match accurately on each landmark;what＇s more,the trained model depends heavily on the clutter background.In view of the drawbacks mentioned above,firstly,we propose the local texture model in clutter background;secondly,we design the 2-D model for corner points;and lastly,we add fringe restrain in the model of partial grey level,so that the points with strong fringe information could have more possibility to become the perfect candidate points.Experiment results show that the improved algorithm is superior to those available in accuracy and robustness.%针对传统ASM方法进行人脸面部特征定位的一些缺点,提出了几点改进算法。传统的ASM方法难以在每一个特征点上都取得较好的匹配结果,并且由于背景图像的不确定性,训练得到的模型依赖于训练样本的背景。针对以上问题,首先针对边缘特征点,提出复杂背景下的局部纹理模型;其次,针对轮廓的角点,使用二维模型和二维的特征点匹配算法;最后,在局部灰度模型中加入了边缘约束,使边缘信息较强的点有更大的可能成为最佳候选点。实验结果表明,改进算法与仿真算法比较,准确性和鲁棒性均有明显提高。