Multiclass Test Feature Classifier for Texture Classification Pattern recognition, Test feature classifier, Ill-class problem, Texture classification, Rank feature

Abstract

A new multi-class pattern classifier called ‘Test Feature Classifier’ is presented. It is based on training a recogniser by training samples of binary patterns and voting primitive scores depending on many trained templates called ‘test feature’, which serves as local evaluation of the features. The method is non-metric and does not misclassify any patterns once learned previously. The two-class version of test feature classifier was of high performance for searching textual region in complex images. In this paper, we extend it to handle multi-class problems and apply it for solving ill-class problems in texture classification. We show the performance of the classifier on more than 1000 real images and compare it with a linear distance-based classifier and a non-linear distance-based classifier. The experimental results of both simulations and real applications show that the proposed classifier has better performance than conventional ones.