Reserch On Image Segmentation Applications Of Markov Random Field

Image segmentation is divided the image into a number of regions which havephysical meaning, thus, and same area have the same or similar nature, differentregions with different properties of the process.Image segmentation is one of the key step from image processing to imageanalysis in the process, which is used for motion estimation, object recognition,scene reconstruction and other senior visual information processing on the premise.The Bayesian theory and Markov Random Field (Markov Random Field, MRF)obtained by combining theoretical framework for image processing, with a fewparameters, For example, the model is simple, easy to combine the characteristics ofother methods, it is widely used in the field of image segmentation. In this paper,MRF was used to reserch the following questions:Firstly, research on the MRF in the MAP (Maximum A Posteriori) solvingapprpromble. Because of the conditions iteration (Iterative Conditional Mode, ICM)algorithm in the whole image can only select a coupling coefficient, thesegmentation result is not enough detail in the case, this paper use quadtreedecomposition princip le, which can give the image a number of couplingcoefficientthe based non-uniform MRF coupling coefficient estimation method, theresults shows improved segmentation algorithm has good effect, the regionalconnectivity is good and get a more delicate. segmentation results has strongadaptability.Secondly, the selection of different initial has much enormous impact on theICM algorithm, for this situation, In order to solve this promble,this thesis willbring Simulated Annealing (Simulated Annealing, SA) algorithm into the imagesegmentation method which used in physics method. this method can get globalconvergence results in theoretically, solving the promble which the initial cause, butconsuming much time. In order to solve this promble, this article will difine thepoint which in the image as the stable point and unstable points, get a SA improvedalgorithm which based on the stable point, in the segmentation process, only theunstable points calculated. The results show that the improved algorithm does notaffect the segmentation results, but reducing the computation time greatly andimprove efficiency.Finally, the ICM algorithm would be combined with the SA algorithm, thecoupling coefficient selection and computation time should be consideredsimultaneously. obtain an ICM-SA algorithm which can solve the promble that theinitial value and the coupling coefficient caused. Experimental results show that the segmentation results with self-adaptive and segmentation effect is well.