摘要
现有的人脸识别系统大多基于数码相机等设备获取的二维人脸图像,当目标的姿态或者摄像机的方位发生改变时,往往会造成图像的变形以至于无法识别。而当输入的人脸为三维图像时,可以进行任意的姿态变换,从而实现对目标的识别。因此,可以通过三维人脸重建并进行空间姿态变换的方法实现任意姿态的人脸识别。在对原有三维人脸识别算法研究的基础上提出了更为通用的方法,该方法将三维深度数据与二维RGB数据结合起来,通过空间变换实现同一人脸的多姿态表示,从而建立人脸库,而在测试时只需要输入普通二维图像即可实现人脸识别。实验结果表明,此方法在采集的人脸库上,得到了很好的识别效果。
Most of the existing face recognition systems acquire two-dimensional face image based on digital cameras or other equipments. Yet when the target's posture or the camera orientation changes,it will usually lead to image distortion so that the image can not be recognized. When the input face image is a three-dimensional one,it can be transformed to any posture,thus realizing the identification of the target. Therefore,by applying the method of three-dimensional face reconstruction and space pose transformation,the thesis aims to realize face recognition in free posture. Through the study of the original three-dimensional recognition algorithm,the thesis proposes a more general method,which combines three-dimensional depth data and two-dimensional RGB data to recognize the same face in different postures by three-dimensional rotation transformation,thereby establishing face database. It only needs two-dimensional image to realize face recognition while testing. The experimental results show that this method gets good recognition effect on the collected face database.
出处
《信息技术》
2015年第8期26-30,共5页
Information Technology
基金
国家自然科学基金青年基金项目(61203143)
国家自然科学基金(60874002)
关键词
人脸识别
姿态变换
深度图像
三维重构
face recognition
pose transformation
depth image
three-dimensional reconstruction