一种基于彩色+深度的人脸识别算法

A Face Recognition Algorithm Based on RGB-D

本文提出了一种基于彩色+深度(RGB-D)的人脸识别方法,以提高识别率.首先从Kinect获得一个具有丰富的头部姿势变化、光照变化等不同条件下的彩色+深度(RGB-D)图像,将获取的同一个人在不同条件下的多个图像看做一个图像集;其次将Kinect获得的原始深度数据用于姿态估计和脸区域的自动裁剪.根据估计的姿态将一组脸部图像集分成多个子图像集.对于分类,本文提出了一种基于块的协方差矩阵表示图像模型在黎曼流形上一个子图像集的方法以降维,并使用SVM模型分别学习每个子图像集,然后将所有子图像集的结果相融合得出最终的识别结果.本文所提出的方法已经在包含不同条件下超过5 000幅RGB-D图像数据集中进行了评估.实验结果表明本文算法可实现高达98.84%的识别率.

In this paper, a face recognition method based on color-depth（RGB-D） was proposed to im- prove the recognition rate. First of all, images are acquired under different conditions, such as wide- range head posture changes and illuminations from the Kinect. Then RGB-D images of the same person under different conditions are selected into an image set. Secondly, the original depth data acquired by Kinect was used for pose estimation and automatic cropping of face region and a group of face image set was separated into several sub-image sets. For classification, a block-based covariance matrix was pro- posed to represent a subset image on a Riemannian manifold to decrease dimensions. The SVM models were used to study each sub-image set separately, and then the results of all the sub-image set together to get the final recognition results. The proposed algorithm has been evaluated on the dataset with more than 5 000 RGB-D images obtained in different conditions. The experimental results show that the pro- posed algorithm can achieve 98. 84% recognition rate.