一种肤色定位的人脸检测算法

进行人脸识别前,首先要精确定位出一幅图像中人脸的位置,为了快速定位人脸位置,本文提出一种肤色定位的人脸检测算法。首先将实际获取的彩色图像转换为YCbCr和HSI空间图像,并将Cb和Cr图像中的数值进行四舍五入处理,结合Cb、Cr、H和S的阈值去除大部分背景,再统计当前图像中的彩色部分的Cb和Cr值,分别取最多2个数值来共同确定肤色位置,最后由当前位置的亮度信息图像排除手等纯肤色部分,准确定位人脸。本文算法能提高正检率并降低误检率,有利于人脸检测。

Face detection algorithm based on skin color location

Face detection needs to locate the position of the face in an image before face recognition. In the paper, a face detection algorithm based on skin color location was presented for rapidly locating face position. Firstly, the actual obtained color images were converted into YCbCr and HSI space images. The numerical values of Cb and Cr image were rounded. A lot of spaces of background were eliminated with combining the thresholds of Cb, Cr, H and S. Secondly, the Cb and Cr values of the color part of the present image were counted, not more than two values of Cb and Cr were chosen respectively to jointly confirm skin color location. Finally, the luminance information image of the present location eliminated the pure skin color part such as hands to accurately locate face. The proposed algorithm can improve correct detection rate and reduce false detection rate to be beneficial to face detection.