: One of the most important stages in image processing is segmentation. One approach to image segmentation is based on finding the most likely labeling of pixels. To aim that, it is assumed that the pixels of each segmentable region have the same distribution. In real scenes, neighboring pixels usually have common properties. Such information about real scenes can be useful to find more homogeneous regions and yields to more proper segmentation. In a probabilistic framework, this property can be modeled by Markov random field (MRF) models and by means of its relationship to the Gi distribution, as the prior distribution, can be entered to the considered probabilistic model. In the most unsupervised image segmentation algorithms, because of difficulty of estimating the number of segmentable regions, it is assumed that the number of these regions is known as the prior information. In this thesis, using a Bayesian approach and assuming the number of segmentable regions is unknown, an unsupervised color image segmentation method is studied. Also, as an application of image segmentation algorithms, interpretation of segmented regions of image is investigated, with a Bayesian approach.