基于Edge Boxes和深度学习的非限制条件下人脸检测

Face detection based on Edge Boxes and deep learning under unconstrained condition

针对光线、旋转、遮挡、平移等因素对人脸检测结果产生的干扰,提出一种基于Edge Boxes和深度学习相结合的人脸检测算法。首先采用Edge Boxes算法提取出可能存在人脸的边界框,提取边界框中的图像并调整至合适的大小,作为卷积神经网络的输入,然后利用卷积神经网络对提取出的图像进行特征提取和分类,最后利用非极大抑制算法排除多余人脸检测框,得到人脸的准确位置。该算法应用于LFW和Yale B人脸数据库的检测率分别达到98.7%和98.5%,识别单张人脸的时间均小于0.5 s。实验结果表明,该算法在检测率和检测速率方面较传统算法都有了很大的提高,对于遮挡、光照、旋转等干扰具有更强的鲁棒性。

A face detection algorithm based on Edge Boxes and deep learning is proposed to eliminate the interference from light,rotation,occlusion,translation and other factors to the face detection results. The Edge Boxes algorithm is used to extract the bounding box maybe existing in human face,in which the image is adjusted to the appropriate size,and deemed as the input of the convolution neural network. Extraction and classification of the features in the extracted image are carried out by means of convolution neural network. The non-maximal suppression algorithm is used to exclude the excess face detection boxes to get the exact location of the face. The detection rate of the proposed algorithm can reach up to 98.7% and 98.5% respectively for LFW and Yale B face databases,and the identification time for single face is less than 0.5 s. The experimental results show that the detection rate and detection speed of the algorithm are much higher than those of the traditional algorithms,and it has stronger robustness against interference from occlusion,illumination and rotation.