Research On Image Segmentation Algorithm Based On Geometric Active Contour Model

Beacause of its better algorithm stability, topological adapability, geometric active contour model has being received more and more attention by people who do research on image segmentation. The relative thoery also play a very positive role in enriching and developing image processing method, promoting industrial information procedure of our country. In this paper, we deeply study the existing geometric active contour models and analyze the properties of these models. According to the limitations of them, we propose several modified algorithms.Firstly, to overcome the defect that the SLGS model can not segment images with intensity inhomogeneities, we define a new Sign Pressure Function utilizing the property that the value of image Laplacian in and outside the contour of the object is opposite in sign. Then, we replace the Edge Stop Function of GAC model with the new Sign Pressure Function and propose SLGS model Based on Image Laplacian. Experimental results demonstrate that the proposed model can segment images with intensity inhomogeneities well, and show the advantage of high calculation efficiency when compared with models with same function. Especially, the proposed model is suitable for detecting object whose contour is thin and narrow.Secondly, according to the deficiency that the WLVLS method is low in calculation efficiency, we define local neighbor fitting image with the model that is commonly used to describe images with intensity inhomogeneities. Then define an enrgy functional by using local neighbor fitting image to approch image to be segmented and propose the Local Neighbor Image Fitting model. Experiments fully proved that the proposed model can satisfactorily segment and correct images with intensity inhomogeneities. Additionally, the proposed model has higher calculation efficiency than WKVLS method.Finally, in order to improve the robustness of the LIF model to initial contour, we first construct Global Image Fitting model by defining global fitting image with global image information. Then combine the advantage of both Global Image Fitting model and the LIF model, and construct LIF model Fused with Global Image Information by introducting the global fitting image into the local fitting image of the LIF model. Experiments show that the first model can segment image with intensity homogeneities very well and permit flexible contour initialization. The second model can effectively decrease the sensitivity of the LIF model to contour initialization and simultaneously improve the calculation efficiency of the LIF model.