Research On The Approach Of Multi-Object CT Image Auto Segmentation Based On Fuzzy Entropy

Image segmentation is a crucial step in a wide range of medical image processing systems. It is useful in visualization of the different objects in the image. Most of the image segmentation algorithms produce binary image,or "foreground and background".While these results are acceptable in some image processing applications such as document processing and Optical Character Recognition systems,they are not satisfactory in medical images where anatomical features of several tissues or apparatus,which are present in the images,need to be detected. In the course of studying home and overseas pertinent algorithms, we find that most of the methods applying fast algorithm have the shortcoming of instability of the result while they solve the computations are usually time consuming. In order to solve the contradiction between the improvement of operation speed and instability of the operation result of multi-object image segmentation,two methods are proposed in this dissertation:(1) Multi-object CT image auto segmentation method based on minimal fuzzy entropy: Firstly the values of the exponent parameters of membership functions of fuzzy subsets and the range of the searching thresholdings can be determined by using the iterative approach and the image histogram and then the thresholdings of minimizing the fuzzy entropy are implemented by searching all possible combinations of every thresholding in determinate searching range.For the sake of change of membership of intersection of membership function of two fuzzy sets,the relation of the parameters of primary GBMF membership function is renewedly defined.Experimental results show that when the new parameter gets the proper value,our proposed method gives the best performance ,but how is the most proper value deserves farther study.(2) Multi-object CT image auto segmentation method based on probability partition and maximum fuzzy entropy: Based on the relationship between the fuzzy clustering and probability partition and a necessary condition of having maximum fuzzy entropy ,the probability partition of each part is derived. In the process of searching the thresholding combinations,we should firstly start with searching the thresholding combinations that satisfy probability partition and then optimal combination can be found by looking for the maximum fuzzy entropy from them. The searching range of the parameters of traditional trapezia membership function is limited.On the basis of transform of the qualification which the parameters need satisfy, membership function is renewedly defined in this method, so every pixel has likelihood of belonging to three fuzzy sets.Experimental results demonstrate the qualification of the parameters in this dissertation is more reasonable.On account of mutual control ofεand the number of the thresholding combinations which satisfy condition is introduced,the thresholding combinations which satisfy probability partition is nonexistence is avoided as well as the thresholding combinations which satisfy probability partition is so many that the computations are too time consuming is avoided.The experiment results show that our proposed methods give good performance for CT image segmentation. The search speed is quick and the results of many runs are steady than using genetic algorithm or simulated annealing algorithm and segmentation is more exact.