Research On Image Segmentation Based On Improved Lightning Attachment Procedure Optimization And Multi-dimensional Entropy

Image segmentation is one of the basic steps of image processing and analysis.Threshold segmentation is one of the more widely used segmentation methods.However,when the number of thresholds is large,the traditional threshold search algorithm(such as exhaustive search)is often inefficient due to the range of solution space is large.Lightning Attachment Procedure Optimization(LAPO)is a new type of intelligent optimization algorithm,which has the advantages of strong optimization ability and few parameters.This thesis intends to use the LAPO algorithm to achieve multi-threshold and image threshold segmentation under multi-dimensional histogram.However,the standard LAPO algorithm still has problems such as unstable optimization capabilities.In view of the shortcomings of the standard LAPO algorithm,this thesis proposes four improved schemes and applies it to the multi-threshold segmentation of images based on multi-dimensional entropy.The main work is as follows:1.Aiming at the shortcomings of the standard Lightning Attachment Procedure Optimization(LAPO),this thesis improves it by opposition-based learning strategy,chaos perturbation strategy and Gauss mutation strategy,and proposes the Lightning Attachment Procedure Optimization based on opposition-based learning(OLAPO),Lightning Attachment Procedure Optimization based on chaos perturbation(CLAPO),Lightning Attachment Procedure Optimization based on chaos initialization(CILAPO)Lightning Attachment Procedure Optimization based on Gauss mutation(GLAPO).Experimental results prove that the improved algorithm does well in stability,and also enhances the ability of optimization,and the improved algorithm is tested in the standard function optimization problem2.The standard Lightning Attachment Procedure Optimization and its improved algorithm are applied to non-noise image segmentation.The experimental results show that the optimization ability and stability of the four improved algorithms are enhanced.Among them,the comprehensive performance of OLAPO is better,and it can achieve multi-threshold segmentation with better effect and strong stability.3.The entropy of different dimensions is used as the segmentation criterion,and the experiment of noisy image segmentation is carried out by using the well behaved OLAPO algorithm.The experimental results show that,in general,three-dimensional entropy segmentation method has the best effect on noise suppression,while it can retain more details;one-dimensional entropy segmentation method can retain more details on some images,but it can hardly suppress noise;two-dimensional entropy segmentation method can suppress noise to a certain extent,but compared with three-dimensional entropy method,the retained image details are less.In addition,the image segmentation method based on OLAPO and three dimensional Shannon entropyperforms well in noisy image.Overall,the optimization capabilities and stability of the four improved algorithms have been enhanced,which can better meet the threshold segmentation requirements,especially for the segmentation of noisy images.Among them,the improved LAPO image thresholding segmentation scheme using opposition-based learning strategy has excellent comprehensive performance in all schemes.