Abstract: An improved face verification algorithm is proposed based on the modular 2DPCA and Client Specific Kernel Discriminant Analysis (CSKDA) because of the disadvantage of CSKDA. CSKDA first transforms an image matrix to a vector which caused high dimensionality and computational complexity and not considered the local feature. While the method first extracts the local features with the original images which are divided into modular sub-images, CSKDA is utilized. The discriminant information obtained from the between-class and the within-class scatter matrix are developed. Moreover, client specific subspace can describe the diversity of different faces. Experimental results obtained on XM2VTS and ORL show the effectiveness of the method.

Key words: Client Specific Kernel Discriminant Analysis(CSKDA), modular 2DPCA, feature extraction, face verification

摘要： 针对客户相关的核判别分析(CSKDA)对图像列向量进行处理数据维数大、计算复杂，对图像整体处理没有考虑到局部特征等缺点，提出M2DPCA和CSKDA结合的方法。新方法对二维数据进行分块后采用2DPCA抽取局部特征，施行CSKDA，不仅考虑了类内、类间的差异，而且可以较好地描述不同个体人脸间的差异性。在XM2VTS和ORL人脸库上的实验结果表明，该方法在验证效果上优于CSKDA方法。

关键词: 客户相关的核判别分析, 模块化2DPCA, 特征抽取, 人脸验证

CLC Number: 

• TP391.4