Texture Classification Using Fractal Dimension Improved by Local Binary Patterns

This paper presents a texture analysis method that combines Bouligand-Minkowski fractal dimension and local binary patterns (LBP) method. The LBP approach is used to obtain "pattern images" from an original input image in order to provide new information sources to be exploited by the Bouligand-Minkowski fractal dimension. Two hybrid approaches were proposed and their results are: "FD(Original image + LBP maps)" (97.12% and 63.80%) and "FD(Original image + LBP maps + STD)" (98.20% and 70.80%) for Brodatz and UIUC image databases, respectively. These results demonstrate that the proposed hybrid method provides a high discriminative feature vector for texture classification.