Texture image segmentation based on spectral clustering ensemble via Markov random field

Image segmentation is a fundamental problem in computer vision. Recently, ensemble learning receives more and more attention for its robustness, novelty and stability. Generally there are two problems in ensemble learning. One is the generation of the individuals of ensemble. The other is the consensus function of the individuals. We focus on the second problem. A new consensus function is proposed for texture images segmentation. To the consensus function, the spatial information of image, that means the adjacent pixels belong to the same class with a high probability, are considered via MRF. Expectation Maximum (EM) algorithm is applied to estimate the parameters of the model and converges fast. The experimental results show that the performance of our model is better than SC using Nyström method and the SCE via mixture model proposed by Topchy for image segmentation.