Research On Medical Image Segmentation Algorithm Based On Otsu Method

In fields of image processing, image segmentation is an important part. It has a wide range of applications on medical research. A great many medical image segmentation algorithms have been proposed, and segmentation results were obtained. However, due to the noise, density inhomogeneity, partial volume effects, and density overlap between normal and abnormal tissues in medical images, the segmentation accuracy and robustness of some state-of-the-art methods still have room for improvement. Among them, the Otsu algorithm with its strong theoretical basis and ideal segmentation result has been favored by many scholars. In thesis, based on the in-depth study of many Otsu improved algorithm, some shortcomings of the existing problems, one-dimensional multi threshold segmentation, improved firefly algorithm and two-dimensional multi threshold segmentation were made corresponding analysis and improvement, specific improvement method is described as follows:(1) This thesis makes research on medical images and one-dimensional multi-level threshold Otsu, and considering the medical images has the characteristics of organ, and due to the density of human body, so this article adds the impact factor of distance to algorithm. And in order to accurate edge information, the article takes gray gradient that is used for edge detection as one of the conditions of a decision segmentation result. As threshold segmentation, gray value as one of the most important influence impact factor is the most indispensable. Through the above-mentioned three image feature information factor, distance, gray gradient fusion constructed comprehensive information of gray histogram, and gives them different weights. In order to realize the automatic multi-level threshold segmentation, k-dimensional tree(k-d tree) is introduced as a framework to achieve the goal of fast automatic determine the threshold number, and realize the automatic threshold segmentation.(2) The firefly algorithm is a novel bionic swarm intelligence optimization algorithm. It is inspired by group behavior that the nature of fireflies exchange of information by fluorescence. As for firefly algorithm, it is when it initializes firefly individual randomly that makes the algorithm in the optimization process fall into local optimal early or find the wrong solution. Therefore, when this article initializes the firefly individual, introduced valley value in the 2-d gray histogram. A certain position in the histogram projection down to get a projection circle, then get the valley of the circle, the valley just get is one initial position of a firefly, and so on, initializes all of fireflies, then the article adds the local information into the update function of firefly algorithm to prevent from falling into local optimal.(3) In order to improve the speed and accuracy of the segmentation, this paper extendeded Otsu from one dimension to two dimensional, a new two-dimensional multilevel thresholding Otsu based on improved firefly algorithm is proposed.In this algorithm, this paper introduces reciprocal of the gray difference of the same class as within-class added to the Otsu’s objective function. The value of within-class is greater, indicating that the current segmentation is better. Aiming to one-dimensional multi-level threshold Otsu algorithm, the paper test on the brain, blood vessels and stomach, and two-dimensional entropy threshold segmentation, image TSMO(Two- Stage Multithreshold Otsu), particle swarm optimization algorithm and the Bacterial Foraging algorithm(BF) have been carried on the contrast experiment, the experimental results show that this algorithm has good segmentation effect. Aiming to two-dimension Otsu algorithm, 127 images of the brain were made a comparing test, and the segmentation results were analyzed based on segmentation accuracy and time consumption to ensure that the algorithm is applicable and effective.