H. Yoshino and Y.Yamashita (Japan)
Pattern Recognition, Wiener filter, Kernel method, Inverse problem
Wiener lter is used widely for the inverse problem. From an observed signal, it provides the best restored signal with respect to the squared error averaged over the original sig nal and the noise among linear operators. In this paper, we propose applying the kernel Wiener filter, which en ables us to handle signals non-linearly by mapping signals to the high dimensional space with kernel trick, to the pat tern recognition problem. We regard a pattern as an ob served signal and provide an identical original vector for the patterns belonging to the same class. Finally we clas sify an unknown pattern into the class of which vector is the nearest to the restored signal in the high dimensional space of the original space. In addition, we apply linear approximation to the kernel function to enable the regular ization based on the distance in the observed signal space to enhance its performance of generalization. And also we adjust the value of the kernel function in the high dimen sional original signal space to improve its ability further more.
Important Links:
Go Back
