Features selection in character recognition with random forest classifier

Abstract

Proper image recognition depends on many factors. Features' selection and classifiers are most important ones. In this paper we discuss a number of features and several classifiers. The study is focused on how features' selection affects classifier efficiency with special attention given to random forests. Different construction methods of decision trees are considered. Others classifiers (k nearest neighbors, decision trees and classifier with Mahalanobis distance) were used for efficiency comparison. Lower case letters from Latin alphabet are used in empirical tests of recognition efficiency. © 2011 Springer-Verlag Berlin Heidelberg.