基于双眼独立动态阈值的人脸预处理及识别

Eyes-independent Dynamic Threshold Based Human Faces Preprocessing and Recognition

针对实际环境中采集的人脸图像预处理问题，提出双眼独立动态闽值方法定位双眼。先划分左右眼大致区域，再采用区域像素点百分比选取灰度值当作二值化的阈值、提取左右眼候选位置后，结合人眼位置的一些判别准则进行最终定位截取双眼间条状区域，应用灰度累积直方图方法，定位鼻子与嘴巴区域在ORL库和essex94数据库上比较了区域划分前后的识别效果，实验结果表明采用所提出的预处理方法后，识别效果有比较明显的提升。

Based on eyes-independent dynamic threshold, this paper proposes a preprocessmg method for human face images collected from real environment. First carves out the approximate area of the left and right eye, and then calculates the gray value from the percentage of region pixels, selects the calculated value as the binarization threshold. After exacting candidate locations of left and right eyes, combined with the eye position criterion the both eyes can be finally located. Locate the region of nose and mouth by applying gwayscale cumulative histogram method based on the striped areas between eyes. Compare the recognition results before and after zoning human face in ORL face database and excess94 #face database, the results show that the proposed preprocessing method can significantly improved the effect ofidentify.