核 Fisher 的鉴别方法 (KFDA) 是模式识别中较为突出的提取图像非线性特征的方法。 为了更好地提取掌纹图像的非线性特征, 将KFDA 方法引入到掌纹识别中。首先对掌纹图像做小波变换进行降维,在保留原始图像轮廓信息和特征的基础上, 用核Fisher判决方法进行特征提取并引入零空间的核 Fisher(ZKFDA) 方法解决小样本问题,最后用最小距离分类器 进行掌纹匹配。通过 PolyU 掌纹图像库,实验结果表明,在不同的特征个数下, KFDA 方法比二维 Fisher 准则 (2DFLD)方法识别率高;零空间ZKFDA的平均识别率高于 KFDA, 并且计算量大大减少。在核函数选取上,取 RBF 核函数的识别性能最佳。

Kernel Fisher discriminant analysis (KFDA) method is a more prominent method in pattern recognition to extract non-linear characteristics. Kernel Fisher discriminal analysis was introduced in the palmprint recognition to extract non-linear characteristics. Wavelet transform was used to reduce palmprint image dimension based on retaining the original image information and features. Kernel Fisher discriminant analysis was used to extract features and the null-space KFDA method(ZKFDA) was introduced to solve the problem of small samples. A classifier to palmprint match was used based on minimum distance. Experimental results show that KFDA performs better than two-dimensional FLD(2DFLD) when the principal component numbers are different. ZKFDA performs better than KFDA in the average recognition rate, and computation is significantly decreased. The recognition performance of radial basis function is the best in the selection of kernel functions.