基于不同分块多特征优化融合的人脸识别研究
针对光照、姿态、表情等复杂情形下人脸识别率较低的问题,提出基于不同分块多特征优化融合的人脸识别方法.首先考虑了局部二值模式、局部相位量化特征和小波变换特征.进一步,考虑单一分块算法会使分割线周边信息不能完整提取,从而丢失对人脸识别的有用特征,提出了人脸灰度图像多重分块的方法.最后,采用遗传算法对不同分块多特征进行权值寻优,得到最优权值.在大规模人脸数据集FRGC2.0数据库上进行实验四验证,验证率达到95.31%(FAR0.1%),首选识别率为99.06%,相比于前期文献,该算法能多方位提取人脸特征信息,提高人脸识别率,且所用特征较少....
Saved in:
Published in | 东北大学学报(自然科学版) Vol. 38; no. 3; pp. 310 - 314 |
---|---|
Main Author | |
Format | Journal Article |
Language | Chinese |
Published |
东北大学 信息科学与工程学院,辽宁 沈阳,110819%东北大学 软件学院,辽宁 沈阳,110169
2017
|
Subjects | |
Online Access | Get full text |
ISSN | 1005-3026 |
DOI | 10.3969/j.issn.1005-3026.2017.03.002 |
Cover
Summary: | 针对光照、姿态、表情等复杂情形下人脸识别率较低的问题,提出基于不同分块多特征优化融合的人脸识别方法.首先考虑了局部二值模式、局部相位量化特征和小波变换特征.进一步,考虑单一分块算法会使分割线周边信息不能完整提取,从而丢失对人脸识别的有用特征,提出了人脸灰度图像多重分块的方法.最后,采用遗传算法对不同分块多特征进行权值寻优,得到最优权值.在大规模人脸数据集FRGC2.0数据库上进行实验四验证,验证率达到95.31%(FAR0.1%),首选识别率为99.06%,相比于前期文献,该算法能多方位提取人脸特征信息,提高人脸识别率,且所用特征较少. |
---|---|
Bibliography: | JIA Ming-xing1 , DU Jun-qiang1 , SONG Peng-fei1 , TIAN Shu2(1. School of Information Science & Engineering,Northeastern University,Shenyang 110819,China; 2. School of Software, Northeastern University,Shenyang 110169,China. ) 21-1344/T LBP(local binary pattern); LPQ(local phase quantization); Gabor; optimal weight; multi-scale blocks; face recognition There is still a problem of lowrecognition rate in face recognition when dealing with complex situations such as illumination,pose and facial expression,so a face recognition method based on multi-feature optimization fusion with multi-scale blocks was proposed. Three complementary characters were considered as the features needed,such as local binary pattern( LBP),the local phase quantization( LPQ) and wavelet transform. Further,a method of multiscale blocks of face grayscale image was proposed to consider that the single block algorithm made the surrounding information not be fully extracted and thereby lost useful features for face recognition. Finally,genetic a |
ISSN: | 1005-3026 |
DOI: | 10.3969/j.issn.1005-3026.2017.03.002 |