【摘要】 找到有效的人脸描述方式是单样本人脸识别的一个很重要的问题.论文提出人脸局部区域融合Gabor特征和LGBP特征的一种新方法,它们能提供人脸描述的互补性信息,比只使用一种特征获得了更好的性能.论文先对人脸进行有效的规范化处理,然后对人脸进行Gabor小波变换并采样,并计算采样后幅度图像的局部二值模式(LGBP).通过将人脸划分成不同的区域,融合不同区域内不同Gabor幅度图像的Gabor特征和LGBP特征,对人脸进行分区域识别,最后对不同区域的分类结果按类的总和决定分类.在AR、ORL和FERET上的试验结果表明,在局部遮挡、表情变化、光照和小的姿态变化方面识别率和当前流行的几个算法相比有一定程度的提高.更多还原

【Abstract】 To find an effective way to describe the human face is a key issue for single sample face recognition. A new approach of fusing Gabor feature and LGBP features on local region of face is proposed in this paper,They have complementary information in the face description and get better performance than either alone. This paper first normalized the image of face effectively,Then conducts Gabor wavelet transformation and sampling on human face,calculates the Local Binary Pattern( LGBP) of amplitude image next.Faces are divided into different regions,and Gabor feature and LGBP feature on different magnitude image for different regions are fused to identify classification of subregion of human face. The classification results are obtained by summing up the classification results of different regions lastly. Experimental results on the AR、ORL and FERET face datasets show that the recognition accuracy outperforms several state-of-the-art approaches and demonstrate promising recognition capabilities against variations including expression,illumination,disguise and small pose change.更多还原