Texture Segmentation With Seeded Region Growing In Feature Space By Integrating Boundary Information

In this study, region-based segmentation of textural images is investigated. For this purpose, the seeded region growing algorithm is used in feature space. In order to make an accurate segmentation, it is crucial to appropriately select the initial seed points as well as to decide where to stop the growing procedure. In the first stage, the boundaries between the textures that will guide the growing process are extracted. Then, the initial seed points are selected according to some intra-region similarity and inter-regional distance criteria in the feature space. At the end of the region growing, the smaller regions are merged according to the boundary information to construct the final segmented image. To discriminate between textures, four different features are used. The first three features are the fractal dimension (FD) of original image, contrast-stretched image and top-hat transformed image, respectively. The fourth feature is the entropy which is a parameter obtained from the spatial gray-level co-occurrence matrix of the image. The experimental results are presented for mosaics with different number of textures.