基于光照归一化分块自适应LTP特征的人脸识别

Face Recognition Based on Illumination Normalization and Block- based Adaptive Local Ternary Pattern

针对复杂光照人脸识别的问题,文中提出一种基于光照归一化分块自适应阈值局部三值模式(Adaptive Threshold Local Ternary Pattern,ATLTP)的人脸识别算法。该方法首先对人脸图像进行光照归一化预处理,消除大部分光照影响;然后对处理后的人脸图像进行ATLTP特征提取。为了更有效地表征人脸特征,进一步将ATLTP特征矩阵划分为大小相等的子块,并对各个子块进行ATLTP特征直方图统计,最后将所有子块的直方图连接起来,构成整幅人脸图像的鉴别特征。根据最近邻准则进行分类识别,在Extended Yale B人脸库和CMU PIE人脸库上的实验结果表明,所提算法可以有效提高复杂光照人脸识别的性能。

To solve the problem of face recognition under complex illumination,an effective face recognition method based on illumination normalization and block- based Adaptive Threshold Local Ternary Pattern（ ATLTP） is proposed. It first performs illumination normalization,and eliminates most of the light effects on face images. Then ATLTP features are extracted from the processed face images. To represent the face features effectively,the feature matrix is divided into several units,and the histogram of each unit is computed and combined as facial features. According to the nearest neighbor principle for face recognition,the experiment on Extended Yale B face databases and CMU PIE face databases demonstrates that significant recognition rate can be achieved under the complex illumination conditions by the proposed method.