Optimizing Gabor Filter and Local Binary Patterns for multi-texture classification

In image classification by texture, it is important to maximize the discrimination
between different classes by using an effective descriptor. The objective of this research is a new hybrid approach using state of the art feature extraction methods and improving the classification percentage of optimum filter by combining it with optimized LBP and find low dimensional size of features. The Gabor filter (GF) parameters are processed by the Artificial Bee Colony (ABC) algorithm to select the optimum filter, whereas pertinent features from LBP histogram are obtained using Rough Set Theory (RST) without impacting
its classification rate. The classification implemented on texture classes is obtained from the Brodatz database. The results from the proposed approach show an improvement in the classification accuracy and processing time of k-folded cross- validation Neural Network classifier over the method of LBP with single filter and a reduced processing time of the classifier.