Adaptive Region Merging Approach for Morphological Color Image Segmentation

This study we focus on the morphological-based image segmentation problem, based on the watershed pre-segmentation with color-alone feature. Based on the color mathematical morphology (MM) method, the similarity measure of merging process between neighboring pixels and regions can be performed as a ranking problem. To avoid the creation of a false color and false segmentation, a hybrid-ordering algorithm instead of vectorized and marginal ordering approaches. Unlike other ordering methods firmly the black color as the reference color to sort pixels, it creates the problem that the scope of the distance measurement is not optimum. We present an automatic reference color selection (ARCS) scheme to obtain the segmentation results more line with human visual perception. Based on a new quartile-based region merging approach, the threshold determination reacts less sensitivity to the context variations of images. To evaluate the system, four quantitative indices are utilized to compare it two typical segmentation schemes in the experiments. It shows that the result of the proposed method is robust to segment the image.