Image Segmentation Based On Teaching-Learning-Based Optimization Algorithm

As one of the important processes of image processing technology,image segmentation has always been an important research topic in the field of image processing and computer vision.With the development of the information age,simple image segmentation methods can no longer meet increasing demand of segmentation.As an important method for solving optimization problems,intelligent optimization algorithms are widely used in many fields,such as industry,military affairs and medical treatment.Therefore,researchers try to combine intelligent optimization algorithms with image segmentation to get a better segmentation effect.Teaching-learning-based optimization(TLBO)is an intelligent optimization algorithm inspired by the teaching process,which has attracted extensive attention of researchers because of its simple optimization process and easy implementation.TLBO has obvious advantages in dealing with optimization problems where the optimal value is near the origin,but its optimization ability is limited in dealing with complex problems,so it needs continuous improvement.In order to improve the optimization performance of TLBO and expand its application in the field of image segmentation,the following improvements are proposed:(1)Aiming at the shortcomings of poor population diversity and easy to fall into local search in basic TLBO,an improved NFDR-TLBO algorithm was proposed.The neighborhood topology and fitness-distance-ratio mechanism was introduced in the improved algorithm.It not only redefines the update rules in teaching stage and learning stage,but also increases the information interaction between individuals while ensuring the strong search ability in early stage of the algorithm.The simulation results on 18 benchmark functions show that NFDR-TLBO has obvious advantages in solution accuracy and convergence speed,and the test results on image segmentation also show the effectiveness of the improved algorithm.(2)Inspired by the idea of ensemble optimization,an ensemble EMTLBO algorithm was proposed.According to the characteristics of TLBO and its different variants,the new algorithm integrated basic TLBO,NSTLBO and TLBO-DE through a specific strategy.The new algorithm divided the population into three subgroups,and assigns an optimal algorithm to each subgroup according to the fitness-based and diversity-based metrics and matching mechanism based on ranking for sub-swarms.It makes the advantages and disadvantages of TLBO and its variants complement each other,thereby improving the overall optimization performance of the algorithm.Furthermore,the experimental results on CEC2014 and CEC2017 test suits verify the feasibility and optimization performance of EMTLBO.Finally,the proposed algorithm is extended to optimize the segmentation thresholds of images and the segmentation performances on different benchmark images show that EMTLBO has good performance in most cases.(3)A multi-objective TLBO algorithm KMOTLBO based on knee point was proposed.The new algorithm extended the single-objective TLBO to multi-objective domain and introduces the adaptive knee point strategy into multi-objective TLBO.The adaptive knee point strategy maintains an appropriate ratio of the identified inflection points to all non-dominated solutions in each front by adjusting the neighborhood size of each solution,and accurately locates the local knee points of Pareto fronts in the mixed population of parents and children to promote species diversity.KMOTLBO not only maintains the diversity of the population,but also improves the convergence ability and reduces the possibility of the algorithm falling into the local optimum.Through the simulation test of multi-objective test functions and comparison with other optimization algorithms,the effectiveness and feasibility of KMOTLBO in solving multi-objective optimization problems are verified.In addition,the proposed algorithm also has advantages in dealing with image segmentation problems,which further verifies the effectiveness of the algorithm.