A study on pattern encoding of local binary patterns for texture-based image segmentation

Abstract

Image segmentation is an important technique for image analysis. For image clustering, the homogeneity of pixel features is usually measured using the Euclidean distance. When textures are used as features for clustering, an encoding scheme that can rationally describe the variations of textures in terms of Euclidean distance, which provides effective clustering results. This study discusses on the problem mentioned above, where the local binary pattern (LBP) is employed as features for clustering. A heuristic algorithm is designed for rearranging the LBP codes. The fuzzy c-means algorithm is used as the clustering method. Some images are applied for evaluation and the results are analyzed. Clustering results using our proposed method and the original LBP encoding are compared. Experimental results show that proper arrangement of LBP encoding improves the performance of image segmentation, without modifying the clustering algorithms.