Unsupervised segmentation of textured image using Markov random field in random spatial interaction

In this paper, we propose an unsupervised segmentation algorithm for a texture image, based on the Markov random field (MRF) in random spatial interaction (RSI). The RSI, which is also another random field, has been adopted to distinguish real texture images with small window size. In this paper, the probability density function of RSI is assumed to be the Gaussian MRF, making the extraction of the texture features easy. The proposed textured image segmentation consists of two stages: texture feature extraction and clustering the feature parameters. In the extraction stage, we use the expectation maximization algorithm, which is widely used for incomplete data problems. Then, the extracted texture parameters are clustered by using the k-means algorithm. The experiment shows good segmentation results for both synthetic and various real images.