Pattern recognition in the presence of noise

摘要

In pattern recognition, the CLAFIC (ClAss-Featuring Information Compression) performs very well. However, the original K-L subspace used in CLAFIC does not consider the case where the patterns contain noise. We propose a new method using the relative K-L operator which minimizes the sum of the mean squared error between the original pattern and its approximation, and the mean squared error caused by the noise under the condition that its rank is fixed. The experimental results of recognition in the presence of noise demonstrate that the performance of our method is superior to that of CLAFIC.