A multi-scale texture segmentation method

In order to realize texture image segmentation, a new texture segmentation method is presented. The method fused with Local Binary Pattern (LBP), Gray-level Co-occurrence Matrix (GLCM) and permutation entropy. After processing the texture image by the rotation invariant LBP operator, the LBP image can be obtained. Then co-occurrence matrixes can be calculated in the LBP image, some features which can describe texture can be obtained in these matrixes, and in the original image, some permutation entropies can be calculated in many different directions. After combining those features and permutation entropies, a feature vector can be built. According to this feature vector, the method made the texture segmentation in multiscale. The center of a homogenous texture is analyzed by using features in coarse resolution and its border is detected in finer resolution so as to locate the boundary accurately. Compared with the method based on permutation entropy and gray feature, the presented method shows visible improvements both in segmentation accuracy, and in increasing boundary precision and region harmony.