项目名称: 基于毛孔尺度面部特征的高效人脸识别研究
项目编号: No.61503084
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 其他
项目作者: 李东
作者单位: 广东工业大学
项目金额: 21万元
中文摘要: 最新的研究表明,普通相机摄像头采集的人脸图像(双眼间距大于280像素即可)中记录了大量皮肤毛孔(一般人眼不可观测)可重复定位检测,经过适当特征提取后具有良好的不变性和可区分性。其纹理和几何排列使其不仅在同一人的不同位置不同,且在不同人之间仍然不同。本项目拟将人脸皮肤毛孔作为一种新的生物信息,提取面向大规模识别应用的高效二值毛孔尺度特征,研究基于毛孔尺度下人脸的新特性,进而实现高精度人脸识别方法,解决同卵双胞胎人脸识别,以及局部整形、遮挡、跨姿态、变光照等复杂条件下的人脸识别问题。研究内容包括,适用于人脸毛孔尺度特征提取的图像预处理方法研究、 高效鲁棒人脸毛孔尺度特征点定位检测与二值特征提取方法研究、基于代数几何可优化的高效稳健拟合算法研究。本项目首次提出将皮肤毛孔尺度特征引入人脸识别领域,具有基础性和重要意义。
中文关键词: 人脸识别;毛孔尺度面部特征;特征点定位检测;二值特征提取;同卵双胞胎人脸识别
英文摘要: Recent researches show that the facial images (inter-pupillary distance greater than 280 pixels) captured by common cameras record a large number of skin pores, which can be detected and localized. The pore-scale facial features extracted by suitable methods are robust and distinctive. The relative positions and the textures of neighboring pores make it different from any other positions skin of the same person or any local skin of the others. In this project, we propose to utilize pore-scale facial feature to achieve efficient and high-accuracy face recognition, including face recognition of identical twins, face recognition under variant conditions, such as illuminations, poses, occlusions. In particular, we will focus on the following research problems: robust and distinctive pore-scale binary feature detection and extraction, optimal robust fitting and feature matching, face recognition of identical twins under variant conditions.
英文关键词: Face recognition;Pore-scale facial feature;Keypoint detection;Binary feature extraction;Face recognition of identical twins