Backtracking Search Optimization Algorithm And Its Application In Image Segmentation

Image segmentation which refers to the process of dividing the image into regions with different characteristics and extracting the target of interest, is a key step between image processing and image analysis, also is one of the basic problems of computer vision.Firstly, this thesis reviews the main image segmentation methods, such as edge detection segmentation method, region segmentation method, segmentation method based on graph theory, the energy functional segmentation method, segmentation algorithms based on machine learning, and threshold method which is used in this paper, threshold method’s basic idea is using one or more thresholds to divide a image into several sections in gray level, the gray value of pixels in the same class belongs to the same section. So the threshold’s selection becomes a key problem of threshold method, and is a key factor affecting the segmentation results. Then it introduces several main threshold selection methods, and solve thresholds by the analytic formula. However, it is a difficult mission to use the analytic formula, because the amount of calculating and computational complexity increase exponentially. So the current mainstream approach is regarding a threshold problem based on criterion function as optimization problem with objective function as criterion function.In this way, large number of methods based on swarm optimization algorithm occur, which is used to solve threshold problems. And swarm optimization algorithm is used in this thesis. Then it briefly introduces the situation of swarm optimization algorithm. And leads to the backtracking search optimization algorithm (backtracking search optimization algorithm, BSA), which is a new bionic algorithm, BSA have a simple structure, strong local search and global search ability, and can effectively and quickly solve many kinds of function optimization problems.This thesis regard the multi threshold image segmentation model as a optimization problem, we employ Otsu and Kapur methods to build two target functions, and then use backtracking search optimization algorithm to solve them, leading to segment image. The proposed approach is applied to nature images segmentation in comparison with other multilevel threshold optimized other algorithms. The results show it is feasible to segment image by using backtracking search optimization algorithm based multilevel threshold. Moreover, compared with other algorithms, it is suggested that the proposed approach can bring better performance.