一种基于P稳定局部敏感哈希算法的相似人脸检索系统设计

A similar-face-image-retrieval system design based on a P-stable locality-sensitive Hashing algorithm

针对智能移动终端、移动机器人安防巡检等应用需求,本文提出了一种基于P稳定局部哈希算法的相似人脸检索系统设计。首先,采用基于局部组合二值特征检测图像中的人脸。进而,通过深度自编码神经网络提取人脸特征。最后,基于所提取的图像的人脸区域特征使用稳定分布的局部敏感哈希算法对每幅图像构建高效索引。实验表明,本文所设计的相似人脸检索系统处理一幅图像的时间约400 ms,能满足实际应用需求,且返回检测结果的误检率低于经典AdaBoost算法。

This paper proposes a similar-face-retrieval system based on a P-stable local hashing algorithm to meet the requirements of intelligent mobile terminals and mobile-robot-security inspection applications. First,our system extracts a locally assembled binary feature to detect a human face in a particular image. Subsequently,a deep autoencoding network is used to compute the subject ' s facial features. Finally,a locality-sensitive hashing algorithm based on a P-stable distribution is employed to construct an efficient index for each image according to the facial features. Our test results show that the proposed similar-face-image-retrieval system can process images within approximately 400 ms,thereby meeting the requirements of practical biometric applications. In addition,the false detection rate of the proposed method is considerably low than that of the classical AdaBoost algorithm.