An automatic unsupervised fuzzy method for image segmentation

Starting from the fuzzy intensity-connectedness definition (chi-connectedness) and the related growing mechanism, a new method is here proposed for the unsupervised, automatic, global region segmentation of digital images, hereinafter referred to as the "Automatic Fuzzy Segmentation" (AFS). One of the major advantages lies in the strict and very simple integration between the analysis of topological connectedness and grey level similarities of the pixels belonging to the same region. By overcoming the previous drawback due to the need of some seed points selection, an iterative processing is here developed, able to adapt to the image content. The automatic selection of seed points is driven by intermediate connectedness results which alternates the analysis of inter-region similarities with inter-region separation measurements. The robustness of the method with respect to the three required parameters is discussed. Example cases related to the biomedical application domain are here presented and discussed.