基于知识的人面器官边缘检测及特征抽取

KNOWLEDGE-BASED EDGE DETECTION AND FEATURE EXTRACTION OF HUMAN-FACE ORGANS

为实现复杂图景中人面的识别,需要抽取有效的特征。本文提出对原始图像做多层直方图均衡化预处理,然后采用多阈值二值化抽取初始边缘,再利用知识指导获得较完善的器官轮廓线,从而抽取出特征用于识别。本文以人面嘴部的边缘检测及特征抽取为例,通过实验证明这个方法是可靠的,能够得到期望的有效特征。

For face recognition in a complex background, extracting effective features is a necessary step. We proposed to equalize the original images in regions of different sizes at first, then to detect edges using multi-threshold binarization. Finally, we obtained complete organ contours directed by knowledge to extract needed features for recognition. The method is applied to mouths to test its reliability in edge detection and feature extraction. Experiments show it, is reliable and the extracted features are effective.