Research On Heuristic Algorithms Based Medical Image Threshold Segmentation

Image segmentation is one of the key and difficult points in image processing research,and it has always been a research hotspot.Image segmentation is the basis of higher-level image processing such as image analysis and image recognition.The segmentation results directly affect the correctness of higher-level analysis and understanding.Image processing based technology is used in a wide range of applications.Not only in the field of image scene,but also in the field of imaging principle,there are many problems such as diverse image sources and different image attributes;the purpose of image segmentation is different;there are various of interferences introduced during imaging,storage and transmission which will degrade image quality;The problem poses a huge challenge to image segmentation.Threshold segmentation is one of the main methods of image segmentation.Threshold segmentation is the most important segmentation method in image processing problems in the field of industrial and medical image processing.This paper combines heuristic algorithms to develop a threshold segmentation method for medical images.In this paper,the basic methods of threshold segmentation and the characteristics of heuristic algorithms are briefly introduced.The basic process of threshold segmentation based on heuristic algorithm is analyzed from the perspective of optimization.By improving the threshold segmentation objective function calculation method and the optimization objective function for the medical image segmentation,several threshold segmentation methods based on improved heuristic algorithm are studied as fellow:(1)Analyze the threshold segmentation method based on heuristic algorithm,by summarizing the threshold segmentation into an optimization problem,summarize basic steps of the threshold segmentation method based on heuristic algorithm.Aiming at the problem that during the traditional heuristic algorithm solving the threshold segmentation problem,the algorithm and parameter determination have high dependence on the specific segmented image and the prior knowledge of the applied target,we present a hyper-heuristic structure oriented to image threshold segmentation.(2)Analyze various of the factors affecting the computational complexity of traditional Otsu threshold segmentation,and propose a comprehensive improvement scheme for these influencing factors analyzed.Use the idea of transforming multidimensional computing into multiple one-dimensional computing in dynamic programming to change multi-threshold segmentation becomes multiple single-threshold segmentation,optimize the underlying computational efficiency of the algorithm.For the threshold search process,we use the heuristic algorithm to improve the computational efficiency.Combined with the characteristics of the segmentation idea,a multi-group PSO algorithm with individual decision-making process is proposed to improve search efficiency and quality at each threshold calculation effectively.The proposed segmentation algorithm is equivalent to multi-threshold segmentation using the classical Otsu criterion in segmentation quality.It is not only superior to the classical Otsu method in terms of computational complexity index,but also has more advantages than multi-threshold segmentation based on recursive multi-threshold segmentation and genetic algorithm.The most prominent advantage is that its calculation time exhibits a linear trend with the increase of the threshold number,which is very beneficial for applications with high real-time requirements.(3)Although the threshold segmentation method based on two-dimensional or multi-dimensional histogram will improve the effect of image segmentation such as noise interference,but the computational complexity will be further increased,the two-dimensional histogram will be reconstructed into one-dimensional by reconstruction method.A histogram,combined with a heuristic search algorithm to solve the threshold.The reconstructed one-dimensional histogram preserves the advantages of suppressing noise and edge effects in the two-dimensional histogram,and the convenience of simple one-dimensional histogram processing.Using SAPSO algorithm to solve the threshold value reduces the problem that PSO is easy to fall into local optimum,and it shows better robustness for multi-threshold segmentation.The experimental results show that the multi-threshold image segmentation algorithm based on improved histogram and SAPSO not only has better overall performance than Otsu,but also has some advantages compared with HOtsu and PSO-based Otsu.The computational efficiency equivalent to PSO Otsu is equivalent to that of HOtsu.(4)A threshold segmentation based on improved ant colony algorithm is proposed.Aiming at the problem of slow convergence in the early stage of traditional ant colony algorithm,an improved ant colony algorithm is proposed and applied to the threshold solution of Otsu segmentation.When the ant colony is initialized,the ant colony is distributed as evenly as possible in the solution space,so that the ant colony searches for the entire solution space as much as possible;in the process of ant colony algorithm execution,the traditional ant colony algorithm is abandoned.The global transition probability parameter generates a random step size by introducing a Lévy flight mode to control the search range of the ant colony.Compared with the traditional Otsu algorithm and the Otsu segmentation based on the classical ant colony algorithm,the optimal threshold can be searched more efficiently and quickly.(5)A two-dimensional Otsu segmentation method based on multi-group dynamic decision making FA(DBM-FA)is proposed.DBM-FA effectively overcomes the shortcomings of the traditional FA algorithm which is easy to fall into local optimum and slow convergence.It optimizes the problem of unstable search results caused by random distribution in traditional FA,and is closer to the physiological mechanism of fireflies in real nature.The algorithm is in line with the purpose of bionics and has a good effect compared to traditional FA and some improved methods.Aiming at the characteristics of heuristic random search optimization algorithm and the characteristics of two-dimensional Otsu threshold segmentation recursive calculation method,a two-dimensional Otsu threshold segmentation algorithm for near-recursive calculation in random search is proposed.(6)A super heuristic structure algorithm is proposed to explore the subspace by meta heuristic algorithm,continuously eliminate inferior subspace,and reduce the search range to improve the efficiency of metaheuristic algorithm.This method improves the efficiency of the heuristic algorithm by optimizing the search space,rather than improving the search ability of the search algorithm itself,that is,improving the overall search efficiency by effectively using the meta heuristic algorithm to search for resources.Compared with the direct use of meta-heuristic algorithm in the whole solution space search,it improves the use efficiency of computing resources;because of the complete isolation between subspaces,a large function space is divided into multiple small function spaces,one A wide range of optimization problems are transformed into multiple small-scale optimization problems.In a small space,the objective function is relatively simple.The search process can find the best point of this range more effectively and quickly,and more effectively avoid falling into the local best.The calculation between subspaces is completely independent,and the subspace is evaluated at a high level,so the subspace search process is advantageous for parallel computing.