Unsupervised multi-spectral image segmentation using watershed transform and MRF model by integrating multi-cue information

This paper presents a combined, two-block framework for unsupervised image segmentation, which is capable of leveraging the best qualities of the watershed transform and MRF models and taking advantage of multi-cue information. The first block extracts various features that respond to different cues of the image and generates their gradient images. Then the obtained gradient images are combined to form a single-valued gradient surface, whose watershed transform provides over-segmented, but homogeneous image regions. The second block of our algorithm groups together these primitive regions into meaningful object based on an improved MRF model. The proposed algorithm is compared with other traditional methods in segmentation of Brodatz texture mosaics and real multi-spectral image. The satisfying experimental results demonstrate the better performance of our new framework.