Abstract: In order to achieve the automatic unsupervised image segmentation, an algorithm based on the adaptive classification and the weighted MRF is proposed. First, combined with the MDL criterion, the number of image classification under the framework of Markov random fields is computed adaptively. And then, the weighted MRF algorithm is used to expand the option range of the potential function, thus to eliminate the complex calculation of the potential function. Finally, by using ICM algorithm to optimize the model, the segmentation image under MAP criterion is obtained. In the Matlab, test results show that the proposed algorithm is effective, which can correctly calculate the number of classification and effectively reduce the segmentation error.

Key words: weighted MRF, MDL, image segmentation, unsupervised segmentation, ICM, MAP criterion