Fuzzy-based probabilistic relaxation for textured image segmentation

This paper describes a fuzzy-based probabilistic relaxation (FPR) for textured image segmentation. The FPR is developed based on an improvement of the conventional probabilistic relaxation which stops after the first few iterations even when the results are still far from satisfaction. The incapability of further improvement in the conventional probabilistic relaxation is detected by a proposed measure of fuzziness. In our FPR, probabilities in the relaxation are suitably adjusted/fuzzified based on a membership function to remove their crisp property such that further improvement can proceed. Experimental results indicate that the fuzzy-based probabilistic relaxation significantly improves the relaxation quality, especially for the textured images composed of components of significantly different sizes. Comparisons with conventional relaxation have also been conducted.<>