Texture-Based Segmentation Using Image Fidelity Indexes

Texture analysis and segmentation is one of the first steps in modeling human visual system for machines. Through the use of texture descriptors, as Gray-Level Co-Occurrence Matrix (GLCM), Local Binary Pattern (LBP) or Sum and Difference Histograms (SDH), this work applies image fidelity indexes to segment the textures of an image. This methodology was evaluated using a dataset of 1,452 images divided into 6 classes of different textures (242 images per class). It was also used to improve the segmentation results of Quadtree decomposition, where the results achieved were always better than the classical version. In another experiment, our proposal was applied to the segmentation of 18 texture mosaic images created from Brodatz textures. All the experiments produced satisfactory results, confirming the robustness of the approach.