摘要
为了解决随着人脸姿态的变化,人脸识别率迅速下降的问题,提出了利用仿射变换和成像原理相结合对待识别的多姿态人脸图像进行姿态调整,将其调整为近似于正面人脸的方法,该方法能够有效地将45°范围内的多姿态人脸图像调整为正面人脸图像。同时,使用改进的SURF(加速鲁棒特征算法)算法对校正后的人脸图像进行识别。在FERET等人脸库及拍摄的人脸图像上进行实验,实验结果表明该方法能够在一定程度上克服姿态变化带来的影响,使平均识别率最高可提高7.0%左右。
As the changing of the face pose, the rate of the face recognition becomes small. Object to the problem, the method which based on affine transformation and image-forming principle corrected the test face images pose to virtual the correspon- ding font face images. The method could correct the multi-pose face images which angle were among 45~. And the face recog- nition based on improved SURF （speeded up robust features） got better effect. The experiment shows that the method over- comes the effect of the pose-changing to some degree and the recognition rate is increased by 7.0%.
出处
《计算机应用研究》
CSCD
北大核心
2014年第4期1215-1219,1228,共6页
Application Research of Computers
基金
国家自然科学基金资助项目(61070245)
关键词
姿态变化
仿射变换
人脸矫正
人脸识别
多姿态
SURF
pose-changing
affine transformation
face correction
face recognition
multi pose
SURF
