Research Of Image Segmentation Algorithm And Its Application

Image segmentation is an essential part of the image processing and computer vision. So far, there is a variety of segmentation methods, and among them active contour model based on variational partial differential equation is a popular one, it is widely studied and used in many applications. The fundamental idea of active contour model is to construct some constraints functions to the segmentation object and realize the segmentation process by curve evolution. According to the property of constraints, the methods basically fall into two categories: edge-based?region-based active contour models.The edge based active contour model, such as the earliest snake-active contour model and some other improved algorithms based on it. But these kinds of methods are sensitive to noise and difficult to detect weak edges. When the initial contour is far away from the target boundary, it is difficult for the edge based model to find the target. Region based active contour model uses region information to define the energy that drives the evolution of curve. One of the most classic region active contour model are regional active contour model without sides proposed by the Chan and Vese, which utilizes global gray information within the internal and external contours. These kinds of methods basically overcome the disadvantages of edge based active contour models, and the segmentation result can be obtained for the noise image and the weak boundary image.The most difficult problem overcoming during the process of image segmentation for variational level set active contour model is that its calculation is much larger and the convergence speed is too low. For this reason,this paper propose a new kind of improved algorithm active contour model based on the basis of previous studies of active contour model. The specific is as following:1. Proposes a new region-based hybrid nonconvex regularization active contour model based on some region-based active contour model. This model constructs a new energy functional which incorporates the local binary fitting model having the property of local clustering of an image and geodesic active contour model. By adding a nonconvex regularization term, the convergence speed of the contour curve is faster, and can well preserve the shape of the region and protect the edge from over smoothing. Then, the minimum of the energy functional will be obtained by the typical finite difference method. Finally, this paper makes simulation experiment on the synthetic and medical images. Simulation results show that the proposed algorithm has faster convergence speed, segmentation accuracy is higher and have better robustness.2. Proposes a new hybrid geodesic region active contour model of local entropy. The model constructs a new energy functional, it introduces a mollifying kernel function to be window function, constructs a new signed pressure force function to replace the geodesic edge stopping function at the same time, and uses local entropy as weight of image fitting energy, then add a nonconvex regularization term to constrain the level set function. The algorithm getting from this not only accelerates the rate of convergence of the contour curve, but also can address image segmentation inaccuracy for image intensity inhomogeneity or blurring caused by the change of illumination or other external factors. At last, do simulation experiments about the algorithm on synthetic and real images. Experimental results show that the proposed algorithm has faster convergence speed, better segmentation accuracy, and is not sensitive to the position of initialization profile curve and has good robustness at the same time.Finally, this paper makes a summary of the work having done, and proposes several directions can go on study for further study of this type of algorithm.