Medical Image Segmentation Based On Improved Geometric Active Contour Model

Image segmentation plays a very important role in digital image processing, the qualityof the segmentation will directly influence the accuracy of subsequent image processing steps,there are many kinds of medical image to handle, such as MRI(magnetic resonance imaging),CT and X-ray image and so on. Because of medical image has the characteristics of thecomplex topology changing, and the geometric active contour model method has theadvantage of processing of these problems, so the geometric active contour is used to segmentimage in this paper.Because of the characters of low contrast, low sharpness, intensity inhomogeneity andpresence of noise appeared in medical image, the final results of the original GAC model cannot reach an ideal effect, so in this paper, we analyze and make some improvements to pointagainst the limitations, the key research findings are as follows:First we make an improvement on the basic of the geometric active contour, westructure a SPF(sign pressure function) through introducing a local area information, then theSPF is used to replace the boundary stop function which is used in the geometric activecontour model, and we get the new model by the evolution of level set method, comparedwith the segmentation result of the traditional GAC method, we can obviously find the newmethod get an better effect, and it has a larger increase both on the speed and accuracy.In the traditional CV model, the global information which can not handle segmentationof minor part well，it is only used while the LBF(local Binary Fitting) model adopts a gausskernel to weight the local area gray, the initialization is not flexible and the contour is easilyto fall into local extremum, so in this paper we combine the advantages of the two methods, toconstitute a new energy function which exploit both the local fitting and global fitting, and weget the result by minimizing the difference between the original images and the fitting images,this method solves the problems mentioned above effectively and has a more robustinitialization. Experimental results verify the feasibility.