A Markov Random Field description of fuzzy color segmentation

Image segmentation is a fundamental task in many computer vision applications. In this paper, we describe a new unsupervised color image segmentation algorithm, which exploits the color characteristics of the image. The introduced system is based on a color quantization of the image in the Lab color space using the popular eleven culture colors in order to avoid the well known problem of oversegmentation. To partially overcome the problem of highlight and shadows in the image, which is one of the main aspect affecting the performance of color segmentation systems, the proposed approach uses a fuzzy classifier trained on an ad-hoc designed dataset. A Markov Random Field description of the full algorithm is moreover provided which helps to remove resilient errors trough the use of an iterative strategy. The experimantal results show the good performance of the proposed approach which is comparable to state of the art systems even if based only on the color information of the image.