利用变形模板提取嘴部特征的算法

嘴部特征提取是人脸识别和人机交互等领域的重要步骤.作者提出了一种改进的利用变形模板提取嘴部特征的算法.针对变形模板对初始位置敏感和匹配易陷于局部极小等问题,首先确定嘴部区域的边框,然后在此区域内仅用边缘和灰度信息,进行全局搜索确定变形模板的最佳参数.算法用两种模板匹配张嘴和闭嘴的情况.实验表明,对于简单背景的灰度图像,该算法可较好地提取嘴部特征,计算速度较快.

An Algorithm of Extracting Mouth Features from Image Using Deformable Templates

Mouth features extracting is a key job in the fields such as lip reading, face recognition, man machine interaction and so on. This paper proposes an improved algorithm to extract mouth features automatically by using deformable templates. This algorithm locates the mouth region at first by rectangular template matching, and then does the deformable template matching in the region using only edge and intensity energies, and a new matching process is adopted. For closing and opening mouth we take two types of deformable templates. The test results show that an exact position of mouth features can be located for the simple background facial images by the algorithm.