An Image Segmentation Method Based On Bayesian And Level Set

Nowadays with the rapid development of computer vision, more and more researchers have been focused on the object tracking, recognition and image processing technology. The image segmentation as an image processing important and crucial step, it is the basis for computer vision. But the kind of image segmentation based on complex to be difficult to unity, according to the characteristics of different areas of the image, the accuracy requirements of the target segmentation algorithm running high and time is short, there will be lots of different segmentation methods. The level set method which based on active contour model, the curve is embedded in a high dimensional space by solving one-dimensional surface of a high minimum level set function of the energy equation, in order to segment the images. During the evolution of level set method, the topology of the curve can be arbitrarily changed, and every iteration of the algorithm steps simple, segmentation result is more accurate than traditional algorithms which are widely used in image segmentation.Bayesian model was introduced to the level set energy equation long before, the core idea of Bayesian probability model is a combination of the likelihood function a priori information and samples, this will solve the problem of that the only prior information bring deviation and avoid errors when there only a sample distribution experimental results show that these two antennas have good broadband performance. The paper proposes an improved algorithm, first of all, the area of the information of the image is added to the original arc information only as a model prior knowledge, and because the information is considered only if the a priori information arc, then the process will lead to segmentation evolution curve because the target has not yet jagged borders and stop the evolution reaches the target area. While only the a priori information contained in the area information of the image, can not guarantee the smoothness of the curve is divided, and therefore a combination of the arc length and area information of the image to improve the accuracy of the image segmentation, the evolution of the curve closer to the true edges of the target. Bayesian probability model while changing only a single Gaussian model consisting of a maximum likelihood function, introduced the Gaussian mixture model, if the target and background are Gaussian model will inevitably arisetarget and background error checking case and if Gaussian mixture model is introduced to avoid the error detection rate caused by a single Gaussian model background and objectives, greatly improving the efficiency of detection and segmentation algorithm accuracy goals. Finally, the combination of these two methods to improve the area of information and information arc formed by combining prior information and background and objectives are subject to two Gaussian mixture model combining experimental results show that the improved greatly improving the image segmentation process precision rate, and for some distribution is more complex, more twists and turns of the target boundary image segmentation better than the level set image segmentation algorithm based on MAP model.