Texture Classification Using Dominant Neighborhood Structure

This paper proposes a new approach to extract global image features for the purpose of texture classification. The proposed texture features are obtained by generating an estimated global map representing the measured intensity similarity between any given image pixel and its surrounding neighbors within a certain window. The intensity similarity map is an average representation of the texture-image dominant neighborhood similarity. The estimated dominant neighborhood similarity is robust to noise and referred to as image dominant neighborhood structure. The global rotation-invariant features are then extracted from the generated image dominant neighborhood structure. Features obtained from the local binary patterns (LBPs) are then extracted in order to supply additional local texture features to the generated features from the dominant neighborhood structure. Both features complement each other. The experimental results on representative texture databases show that the proposed method is robust to noise and can achieve significant improvement in terms of the obtained classification accuracy in comparison to the LBP method. In addition, the method classification accuracy is comparable to the two recent LBP extensions: dominant LBP and completed LBP.