Gvf Snake Model-based Contour Extraction Method

As the key step from image processing to image analysis, contour extraction lays the foundation for feature extraction, target recognition and classification, understanding, etc. And it can be used extensively in many fields such as biology medical image processing, virtual reality, autonomous guided vehicles, industry online automatic checking.Contour extraction based on active contour model (ACM), which is one of the novel and efficient extraction ways, shows great advantage over the traditional extraction algorithm. However, active contour model also has some disadvantages. This thesis systematically analyses existing methods of contour extraction. On this basis, it puts emphasis on contour extraction based on active contour model. And aiming at the shortages of GVF Snake model, it proposes the improved methods. The main work and conclusions in this thesis are as follows:(1) Summarize existing methods of contour extraction, mainly focusing on active contour model.(2) Illustrate principles of parametric active contour model, such as its mathematical model, working mechanism, basic properties, as well as three main solutions to energy minimization and their merits and demerits.(3) Elaborate principles of gradient vector flow and its numerical implementation method. And in comparison with traditional snake model by experimental analysis, GVF Snake model has better properties, such as larger capture range of external force, convergence to boundary concavities.(4) Discuss the distribution characteristic of GVF external forces and the effects of initial contours to contour extraction. On this basis, the watershed transform is adopted to obtain the initial contour of GVF Snake. We use marker-based watershed to solve the over-segmentation problem. Meanwhile, watershed is computed using the gradient vector flow instead of the classical gradient. The main processes are summarized as follows: Firstly, compute gradient vector flow of the image, and use it as altitude of watershed transform. Secondly, make markers inside and outside the object boundary, and segment the gradient vector flow image using marker-based watershed. Finally, GVF Snake model converges to the precise edge from the initial contour given by the improved watershed transform. Experiments show that the initial contour obtained by this method is close to the real object boundary and lays the foundation for GVF Snake extracting objects accurately.(5) Combined GVF snake model with graph cuts theory, a fast concave object extraction algorithm is proposed. On one hand, enough GVF external forces iteration is needed in order to make active contour move towards to the object boundary. On the other hand, graph cuts algorithm can extract non-concavity objects quickly though it can not extract objects with higher curvature. Therefore, incorporate the advantages of both GVF Snake and graph cuts to extract the concavity object boundary. Use graph cuts algorithm to implement extraction of non-concave part of the object boundary. Then use GVF Snake model to implement extraction of concave part of the object boundary. The main processes are summarized as follows: Firstly, give an initial contour CO, dilate current contour into its contour neighborhood, convert the contour neighborhood into a network, and compute the minimum cut to obtain a new contour. Secondly, let the new contour be current contour, repeat above steps until a resulting contour C1 reoccurs. Finally, contour C1 is input to the GVF Snake model and begins its evolvement to the interested object boundary C2. Experimental results show that the algorithm can converge to concave object boundary fast and exactly.