In this paper we present an unsupervised color image segmentation algorithm using the information criteria and a fuzzy theory. We propose this method to estimate the number of color image clusters and the optimal radius associated with minimizing the value of the proposed criteria. The experimental results demonstrate that this approach compresses the image in a small number of clusters without losing the informational contents of the image and we reduce the number of parameters using the process of segmentation, we also decrease the computational time. The color image segmentation system has been tested on some usual color images; "House", "Lena", "Monarch" and "Peppers".