随着深度学习技术的发展,人脸识别在受控环境下的准确率已经达到了非常理想的效果.然而,真实环境下获取的人脸图像往往因为遮挡而难以识别.针对遮挡条件下的人脸识别准确率不高、稳定性差的问题,结合传统的人脸分块和深度卷积神经网络,提出一种基于分块的有遮挡人脸识别算法.基于人脸特征点定位的结果进行人脸分块,使用一种改进的轻量级卷积神经网络进行各个人脸区块的特征提取;利用多分类网络结合输入区块的额外信息进行人脸区块的遮挡判别;结合人脸块特征与遮挡二分类判别结果获取表征遮挡人脸的特征.实验结果表明,经过以上步骤提取出的特征对遮挡具有较强的鲁棒性,并且在满足一定的条件下,即使人脸由大面积遮挡也能在实验数据集上保持94%的准确率.%With the development of deep learning technology, the accuracy of face recognition in controlled environments has achieved very good results.However, face images acquired in real environment are often difficult to identify due to occlusion.Aiming at the problem of low accuracy and poor stability of face recognition under occlusion condition,an algorithm of face recognition with occlusion based on partition were proposed combining with traditional face segmentation and depth convolution neural network.Firstly,face segmentation was done based on the result of face feature point location.An improved lightweight convolution neural network was used to extract the feature of each face. Then the multi-classification network combined the additional information of the input block to distinguish the occlusion discrimination of the face block.In the end,the characterization of occlusion of human face was obtained by combining the results of human face block and occlusion dichotomy.The experimental results showed that the features extracted from the above steps were robust to occlusion,and the accuracy of 94%could be maintained on the experimental dataset even if the face was covered by a large area under certain conditions.
展开▼
