Abstract: In this paper, a new method for face description and recognition is proposed, which is based on feature fusion of Gabor and Local Ternary Pattern（LTP）, and in this new algorithm, the paper alse improves the local ternary pattern and presents a new Local Adaptive Ternary Pattern with an adaptive threshold（LATP）. First, the normalized face image is decomposed by multi-scale and multi-orientation Gabor filters to extract a series of Gabor magnitude maps. Then, Local Adaptive Ternary Pattern（LATP） is used to operate on each extracted Gabor magnitude map to extract the local neighbor pattern and the face image is described by the weighted histogram sequence of all extracted local neighbor patterns, and the weight is determined by information entropy. Finally, the histograms of train images and test images are matched by the chi-square distance. The experiment is conducted on the ORL and Yale face databases. The test results show that the new method has better adaptability under expression changes and illumination variations and has more strong robustness under noise jamming.

Key words: face recognition, Gabor filters, local ternary pattern, adaptive threshold

摘要： 提出一种融合Gabor特征和局部三值模式（LTP）的人脸识别方法，并在算法中对局部三值模式（LTP）进行改进，提出能够自适应阈值的LATP算子。对归一化后的人脸图像进行多尺度、多方向的Gabor滤波提取其对应的幅值特征，在每个幅值图像上进行LATP运算，抽取局部邻域关系模式，这些模式的区域直方图再经过信息熵加权并串联得到最终的人脸描述，识别过程使用[χ2]距离对特征直方图进行相似度匹配。在ORL和Yale人脸数据库上实验，结果表明提出的算法对人脸表情和光照变化具有更好的适应性，对噪声干扰具有更强的鲁棒性。

关键词: 人脸识别, Gabor滤波器, 局部三值模式, 自适应阈值算子