摘要
针对人脸识别中的遮挡问题,提出了一种多层重建人脸算法框架。每层利用上层输出的训练图片对待测图像进行重建,并针对重建图片从训练集中筛选出若干最近邻图片,作为新的训练集输出给下一层,最后一层输出人脸识别结果。在多层重建之前,通过预筛选进行优化。实验结果表明,两层迭代主成分分析法(principal components analysis,PCA)和三层缺口PCA分别比迭代PCA和缺口PCA识别相对错误率下降12%和44%,识别时间减少了51%和72%,验证了该方法的有效性。
In view of the occlusion problem in face recognition,a multi-layer reconstruction algorithm frame is proposed. The upper output of the training image is used in each layer to reconstruct the image,and some of the nearest neighbor images are filtered out from the training set as the new training set output to the next layer. The final layer outputs face recognition results. Before the multilayer reconstruction,the optimization is made by preselection. The experimental results show that the three layer PCA and the three layer gap PCA are better than the iterative PCA and the gap PCA,with the error rate decreased by 10% and 43% respectively,and the time reduced by 51% and 72%.
出处
《北京信息科技大学学报(自然科学版)》
2015年第5期30-35,共6页
Journal of Beijing Information Science and Technology University
基金
北京市属高等学校创新团队建设与教师职业发展计划基金项目(IDHT20130519)
关键词
人脸识别
人脸重建
面部遮挡
face recognition
face reconstruction
face occlusion