Abstract: This paper proposes a method which combines global features of Two-Dimensional Principal Component Analysis（2DPCA） and Two-Dimensional Fisher Linear Discriminate Analysis（2DFLD） with local feature of Local Binary Pattern（LBP）, and applies them in multi-modal face recognition based on 2D intensity image and 3D depth image. Then the normalized similarity scores of different methods will be fused by weighted sum rule. Experimental results show that LBP will improve the performance of 2DPCA and 2DFLD, and score fusion of 2D intensity images and 3D depth images makes improvement too, the highest recognition rate on CASIA3D database has achieved 94.68%.

Key words: global feature, local feature, intensity image, depth image, score fusion

摘要： 提出将全局特征表征方法2DFLD、2DPCA与局部特征表征方法LBP相结合，应用到人脸二维强度图和三维深度图进行识别；对不同分类方法的识别得分再进行归一化加权融合。对比实验结果表明，LBP对2DFLD和2DPCA的识别结果有改善作用，二维强度图和三维深度图的得分归一化加权融合对整个识别率也有一定的改善，在CASIA3D人脸数据库上的识别率最高可达[94.68%]。

关键词: 全局特征, 局部特征, 强度图, 深度图, 得分融合