Research On Image Multi-threshold Segmentation Technology Based On New Heuristic Search Algorithm

Lots of novel heuristic search algorithms have been proposed with the increasement of natural phenomenon and artificial life in recent years,such as Fireworks Algorithm(FWA),Chicken Swarm Optimization(CSO)Algorithm and Extension Artificial Raindrop Algorithm(ARAE)and so on.Their applications on image segmentation are seldom reported at present.Focusing on the heuristic search algorithms presented since 2010,this dissertation analyzes the ideas of those methods,and tries to introduce them to image multi-thresholding.The main innovations of this dissertation include the following aspects:(1)Aiming at the improvement of FWA performance,the procedure of firework explosion is first deep analyzed.Then,the explosion operator is improved to accelerate the convergence speed of the FWA.Hence,an image multi-thresholding method based on the IFWA is proposed.In the method,the three-level stationary wavelet transform is employed to denoising the original image,and then the Otsu method is adopted as the fitness function to evaluate the quality of the individuals.Finally,IFWA finds out the best threshold.Experimental results indicate that the IFWA significantly accelerates the convergence speed and reduces the segmentation time when compared with the FWA.(2)For the CSO algorithm,we analyzed the behaviors of the rooster,the hen and the chick in three roles,and put forward an adaptive image multi-thresholding method based on the CSO algorithm.Firstly,the threshold number is determined adaptively by the histogram of the image,and then the Otsu method is adopted as the fitness function,through the mutual cooperation of the three roles and the iteration of the method to search the optimal threshold.The experimental results show that the method is better than the image multi-thresholding methods based on the Shuffled Frog Leaping Algorithm(SFLMT)and the Artificial Fish Swarm Algorithm(AFSMT)in the three aspects of convergence speed,stability and segmentation effect.(3)Aiming at the ARAE,we put forward a kind of image multi-thresholding method based on the ARAE by deep analysis of the five basic changes of raindrops.This method regards the Otsu method as the fitness function,and the optimal threshold is determined by the collaboration of multiple raindrop subgroups.Our experimental results show that the stability and convergence speed of this method is better than that of the methods based on the SFLMT and the AFSMT.