On Combining Region-Growing with Non-Parametric Clustering for Color Image Segmentation

Region-based and clustering-based techniques are two of the most important segmentation methods, and both of them have their advantages and disadvantages. In this paper, we present a color image segmentation method combining region-growing with non-parametric clustering technique. First, a bottom-up region-merging technique is used to yield an intermediate result. This procedure takes into account simultaneously the spectral properties of pixels as well as their spatial information, which is not fully utilized in clustering technique. Second, a clustering technique based on mean shift algorithm is used to cluster similar image objects in the intermediate result. In the mean shift procedure, we adopt adaptive bandwidths instead of a single one over the entire feature space. The two steps of image segmentation are performed in an unsupervised way. The validity of the proposed method is verified on various color images.