基于Fisher判别分析的加权估计纹理分析  

Weighting Estimation for Texture Analysis Based on Fisher Discriminative Analysis

作　　者：从继成[1] 张韧志[1] 

机构地区：[1]黄淮学院,河南驻马店463000

出　　处：《实验室研究与探索》2015年第2期24-28,共5页Research and Exploration In Laboratory

基　　金：河南省教育厅科学技术研究重点项目(13A520786)

摘　　要：传统的纹理分析方法仅以每个脸部区域的相对贡献来标记全局相似度。针对这种以局部表示全局而导致不能很好地进行特征提取的问题,提出了基于Fisher判别分析的加权估计纹理分析方法。首先使用局部二值模式或者局部相位量化对图像进行纹理编码,然后将其划分成各个大小相等且不重叠的局部小块,从相似空间中提取出最具识别力的坐标轴,利用Fisher线性判别分析方法对其进行纹理分析,通过权值优化给出了最佳解决方案。最后,在FERET和FEI两大通用人脸数据库上的实验验证了所提方法的有效性。实验结果表明,相比其他文献中提出的纹理方法,所提方法取得了更好的识别性能。Traditional texture analysis methods mark global similarity only by related attribution of each face area. For the issue that global infnrmation is represented by local information which causes bad feature extracting, weighting estimation for texture analysis （WETA） based on Fisher discriminative analysis （FDA） is proposed. Firstly, face images are divided into some non-overlapping local patches with same sizes after texture coding by using local binary patterns （LBP） or Ioeal phase quantization （LPQ）. The solution is given by the most dis.riminatiw axis within a similarity space using Fisher discriminative analysis and weight optimization after extracting coordinate axes with the ,hOSt discrimination. Finally, the efficiency of proposed method is verified by experiments condueted on the FERET and on the FEI face databases. The experiments indicate that the proposed method brings a better recognition performance in comparison to othe, weighting methods proposed in the literature.

关 键 词：人脸识别 FISHER判别分析 加权估计 纹理编码 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]