The Use of Gradient Based Features for Woven Fabric Images Classification

Abstract

Texture classification is used in various pattern recognition applications that possess feature-liked appearance. One of the main texture types is the woven fabric texture. This paper aims to improve the classification accuracy of this type of texture based on extracting a directional based texture features. Three different types of features are proposed: (i) first order gradient feature vector, (ii) max-min gradient feature vector, and (iii) second order gradient feature vector. Each one of these feature vectors is studied individually, and then the possible combinations of them are studied also. This study applied on 22 classes of woven fabric with 225 images per class taken from the Brodatz album. The experiments showed that the results are competitive to that gotten from the other popular methods in this field, such as GLCM, Gabor filters, wavelets and other transformation methods. The test results indicated that the attained average accuracy of classification is improved up to (99.909%) for the training set and (99.714%) for the testing set.