Bayesian Image Segmentation using HMRF Algorithm
Image segmentation is a process of dividing the image in to some distinct regions. These regions shave specially coherent in nature and have similar attributes. This technique is widely used for image analyses and to interpret the desired feature. In this present paper, we will study about the hidden Markov random fields (HMRF) and find its expectation maximization algorithm. The main idea behind developing HMRF is to adjoin the “data faithfulness” and “model smoothness” that show very similar nature with the active contours, GVF, graph cuts, and random walks. Here we also use the HMRF-EM along with the Gaussian mixture models, and then we use it on color image segmentation process. These algorithms are implemented in MATLAB. In color image segmentation experiments, we observe that the result obtain from HMRF segmentation are much smoother then the direct k-means clustering. The segmented object is much closer to the original shape than clustering. The segmentation time for Bacteria 1, Bacteria 2, SAR and brain images are 0.35, 0.43, 0.12 and 0.12, respectively. The accuracy for Bacteria, Bacteria 2, SAR and brain images are 97.70, 98.06, 98.89 and 97.35%, respectively.
Keywords: Bayesian methods, convex optimization, image segmentation, spatial mixture models, Potts Markov random field
- There are currently no refbacks.