Image Segmentation Method Based On Active Contour Model

Image segmentation is a fundamental and key issue in image analysis and pattern recognition. So far, researchers have proposed hundred image segmentation algorithms.The active contour model had become the hot issue because of its better stability and accuracy of segmentation, and it had been used in medical image segmentation, edge detection and target tracking, etc.This paper focuses on the geometric active contour model, the main work is as follow:The geometric active contour model based on boundary relies on the condition which stopping the curve evolution depends on the gradient information. When the image holds weak edge and strong noise, the evoluting curve may exceed the boundary of object, as a result, it could not segment the object. The C-V model based on global gray statistical information has the disadvantage which could not segment precisely the medical ultrasonic and magnetic resonance image with uneven gray. This paper proposed a geometric active contour model based on local polarity information, the local polarity information can measure the gradient vector direction, increase the influence to contour by local pixels, and decrease the influence by further pixels, it can distinguish effectively that the pixel locate on the boundary or not. The proposed method has a better segmentation result for neterogeny image. In addition, introducing a signed distance function constraints items to the level set method, The constraints items can keep the level set function closing to signed distance function, which avoids to initialize the level set function again and improves the speed of curve evolution. We used the true and composite image to test the proposed method. The results indicate that the model can segment quickly the uneven medicine image and noise image.GAC is a model which is based on boundary segmentation, it makes use of image gradient information to drive the evolution of the curve, it not only easily leads to the evolution of the curve crossing the border in weak edges and no edge images, but also it is hard to converge to the edge of deep depression and cannot segment multi-loop contour image. Although the C-V model based on the global segmentation takes full advantage of the edge and region information of images, it cannot segment the uneven grayscale target images. For the sake of making full use of image’s edge and the region information, the paper proposed an improved GAC model. First, let SPF function basedon region instead the ESF function in GAC, because SPF can effectively make the evolution curve stop in weak edges, fuzzy edges and recessed edge. Then proposed a selective level method to achieve numerical calculation, the method not only has the ability of choosing part segment, but also can segment the target of interest in high efficiency. Experiments show that the proposed model’s performance in segmentation accuracy and computational complexity aspects are better than GAC and C-V models by comparing the last synthetic images and real images.