基于投影灰度差的遮挡人脸检测算法

针对现有的人脸检测算法,需要样本数量大、训练与学习时间长等缺点,提出了一种基于投影曲线特征的遮挡人脸检测算法。该算法利用了人脸二值化的水平投影,确定人眼区域,减小了数据量和其他面部特征的影响；得到眼部区域二值化的垂直投影,利用投影曲线的特征,对眼部是否有遮挡进行判断。该算法思想简单,无需训练与学习,提出新的判断指标-归一化灰度差有明确的物理意义,有效的提取了眼部区域的特征。仿真结果证明了该算法的有效性,且与同类算法相比,缩短了检测时间,提高了检测率。

Projection Curve Gray Level Difference based Face Occlusion Detection

In order to solve the problem that traditional face detection needs a large number of samples to train and learn,a projection curve based face occlusion detection is presented. It uses gray projection curve of the binary face image to get eye region,and reduces the amount of data and impact of other facial features. The feature of projection curve of the binary eye region is used to determine if there is any occlusion. The method has no need of training and learning, and the normalized gray level difference,which is efficient extraction form eye region information,has clear physical meaning. Experimental results show that the method is effective,and it can increase the speed and improve the accuracy compared with other methods.