一种带遮挡的人脸特征快速提取算法  

作　　者：高健豪 宋春林[1] GAO Jian-hao;SONG Chun-lin

机构地区：[1]同济大学电子与信息工程学院

出　　处：《信息技术与信息化》2019年第12期104-107,共4页Information Technology and Informatization

基　　金：国家科技重大专项资助(项目编号：2017ZX05005001-005)

摘　　要：人脸识别技术是目前发展最快也最具有潜力的生物特征识别技术,具有巨大的发展空间以及应用领域。而在非强制配合脸部图像采集的场景中,脸部特征通常被口罩、墨镜等物品所遮挡,脸部特征信息不完整,无法正确提取脸部特征信息,严重地影响到了最终识别结果。目前已有的针对带遮挡的人脸特征提取算法在准确性和速度性能以及二者的平衡方面具有很大的缺陷,无法满足实际应用需求。因此,本文设计出一种基于视频的带遮挡的人脸特征快速提取方法,在人脸非完全遮挡的情况下,准确地对脸部未遮挡部分特征进行提取。通过仿真实验说明,本文算法具有更低地复杂度以及环境要求,应用场景更加广泛。Face recognition technology is the fastest growing and most promising biometric technology, with huge development space and application fields. In scenes that are not forced to cooperate with facial image acquisition, facial features are usually obscured by articles such as masks and sunglasses, and facial feature information is incomplete, and facial feature information cannot be correctly extracted, which seriously affects the final recognition result. At present, the face feature extraction algorithm for occlusion has great defects in accuracy and speed performance and the balance between the two, which cannot meet the actual needs. Therefore, this paper designs a video-based method for quickly extracting facial features with occlusion. In the case of non-complete occlusion of the face, the features of the unoccluded part of the face are accurately extracted. The simulation experiments show that the proposed algorithm has lower complexity and environmental requirements, and the application scenarios are more extensive.

关 键 词：遮挡人脸 帧间差分 平均脸 ADABOOST 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]