Extraction of noise robust rotation invariant texture features via multichannel filtering

Abstract

An efficient and accurate method of extracting rotation invariant texture features via a multichannel Gabor filtering technique is presented. A main focus of the paper is a thorough investigation into the optimum parameter settings for the method. Experiments include the exploration of different frequency combinations, sampling intervals and number of features. The optimum settings are used to test the method's texture classification abilities on a database of over 1320 images originating from 44 different texture classes. The resistance to noise is measured via the addition of various levels of Gaussian noise to each image before classification.