Research And Improvement Of Medical Image Segmentation Algorithm

For medical images, the image segmentation is a key step in its analysis studies. It is the base to explain the premise of object recognition,3D reconstruction and a series of work, It has been attracting people's attention over the years. But so far, there are still many challenges such as fuzzy boundaries, weakening fracture and other issues,in the algorithms of medical image segmentation. In this paper, the traditional algorithms of medical image segmentation can not meet the requirements for modern medical image processing, This paper proposes an improved segmentation algorithm for medical image processing.Currently,the algorithms of medical image segmentation, studied most by Chinese and foreign scholars, were based on graph cut energy minimization algorithm and the active contour model. Rother, who made grabcut graph cut algorithm,reduces the user interaction.But its accuracy is not high, and has a low degree of automation, which makes it difficult to apply to the real life. On the other hand, active contour model based on a partial differential method can fuse image prior information and has the ability to obtain a continuous smooth contour curve. But it is susceptible to noise and weak edges,which would be trapped in a local optimum caused by leakage of the border. In this paper, in order to improve the accuracy,and the speed of segmentation, this thesis makes an improvement on grabcut algorithm and the traditional level set active contours,and gets better experimental results. The main content and innovation of this study is listed as follows:1?An improved algorithm of medical image segmentation is proposed based on grabcut.This chapter proposes an improved method for medical image segmentation, using the integration grabcut and snake algorithm. Snake model can efficiently combine the prior knowledge of image with a variety of image processing algorithms, but it is susceptible to noise, which will lead to fault segmentation. Firstly, this chapter uses PM model to eliminate small bright noise spot. Secondly,this chapter uses threshold-based segmentation to estimate initial contour of Snake model and then snake model is used to obtain a more accurate object boundaries,which utilized to obtain grabcut initialization rectangle. Finally, the rectangle is passed to grabcut algorithms to extract an accurate real-time medical image contour. In order to demonstrate the improved accuracy of the algorithm, this chapter also introduces the division ratio standard quantitative analysis. Experimental results show that the proposed algorithm combines the advantages of snake algorithm and grabcut algorithm, without human interaction conditions, the average accuracy rate of image contour segmentation can reach to 93.7%, which meets the requirements of medical image analysis.To further reduce the time consumption and to obtain high accuracy cell image segmentation.By setting the threshold, the number of iterations of grabcut algorithm can be determined adaptively. Time-consuming and the overall error rate can get a balance.2?An improved algorithm of medical image segmentation is proposed based on CV and LBF.This chapter proposes a level set algorithm of adaptive weighting factor,which is used to segment image. First, this chapter uses PM model to eliminate small bright noise spot, enhance the image boundary. Then,using an adaptive weighting factor to adjust the weights of CV algorithm and LBF algorithm automatically based on the image information. This design of co can reflect the gray uniformity of the medical image, So that when the image gray is nonuniform, LBF algorithm is stronger. Otherwise,CV algorithm is stronger. And for CT images and MRI images our algorithm can work better than traditional CV algorithm and LBF algorithm.So that can get more accurate segmentation accuracy and higher segmentation efficiency.