Gabor Contrast Patterns: A Novel Framework to Extract Features From Texture Images

Abstract

In this paper, a novel rotation and scale invariant approach for texture classification based on Gabor filters has been proposed. These filters are designed to capture the visual content of the images based on their impulse responses which are sensitive to rotation and scaling in the images. The filter responses are rearranged according to the filter exhibiting the response having largest amplitude, followed by the calculation of patterns after binarizing the responses based on a particular threshold. This threshold is obtained as the average energy of Gabor filter responses at a particular pixel. The binary patterns are converted to decimal numbers, the histograms of which are used as texture features. The proposed features are used to classify the images from two famous texture datasets: Brodatz, CUReT and UMD texture albums. Experiments show that the proposed feature extraction method performs really well when compared with several other state-of-the-art methods considered in this paper and is more robust to noise.