Unsupervised texture segmentation based on multi-scale Local Binary Patterns and FCMs clustering

This paper present an efficient multi-scale approach to unsupervised texture segmentation based on features extracted from Local Binary Pattern (LBP) histograms and fuzzy C-Means clustering with spatial information. In the approach, a multi-scale version of LBP is firstly adopted to overcome the region limitation of basic LBP by extending to larger scales for texture-content extractions. Texture features consisting of averaged intensities, LBP histogram distributions at different scales are then computed within preset windows. Finally, a modified fuzzy C-Means clustering is performed for small region-based segmentation where the spatial position is involved in the object function for enhancing the spatial-dependency among feature vectors within a texture class. The performance of the proposed method is demonstrated on segmentation of several multi-textured images and comparison studies on feature selection analysis are shown on its effectiveness.