和声搜索算法在数字图像分割中的应用研究

发布时间：2019-01-04 20:27

【摘要】：随着科技的快速发展,各个行业领域对数字图像的运用也日益增多,图像分割是图像处理的一种基本技术,因此逐渐受到人们的重视。图像分割方法可以解释为把图像分成特征有差异的部分,再提炼有效的部分出来,且是图像处理过渡到分析的重要途径,在图像领域中具有十分关键的位置,同时在大多数领域应用广泛,取得了较好效果。现如今,越来越多的智能算法被应用于图像分割领域,大多优秀的智能算法渐渐代替了以往常用的方法,在现今已是解决多数优化问题的有效方式。而现在比较常用的智能算法都是由于其本身的某些优秀特性以少量的运算时间来获取较高的成效。近年来,Geem等提出了一种元启发式算法——和声搜索(Harmony Search,HS)算法,且对HS进行了深入研究,并与遗传算法、模拟退火算法和禁忌搜索等进行比较,实验结果显示HS算法在有关问题上性能较好。然而,目前关于HS的研究与应用仍有很多问题亟待解决。本文主要是研究对智能算法进行改进并把改进的智能算法用于提高传统分割算法的效率。所做工作如下:首先,本文阐述了图像分割的几种经典方法和HS的来源、基本原理、具体步骤等,分析了它们的优点与不足,总结了HS在工程应用中所面临的问题及其当前的主要研究方向,并论述了几种经典的改进HS。然后,针对HS存在着易陷入局部最优从而导致提早收敛的不足,提出了一种融合局部搜索的和声搜索(Local Search technique fusion of Harmony Search,LSHS)算法。在LSHS算法中,将最优和声向量与在种群中随机选择的两个和声向量进行线性组合生成一个新的和声,从而扩大了局部搜索区域同时也提高了算法的收敛速度。用9个标准测试函数将本文的LSHS与HS、GHS(Global-best Harmony Search,GHS)进行了测试对比,得出的试验效果显示LSHS的结果更优,性能更好。最后,由于优化方法可以用于寻找最佳阈值,因此本文将提出的LSHS用于最大熵分割。并将HS、GHS和LSHS三种算法对灰度与彩色图像分别采取分割处理且对比实验仿真,结果显示LSHS最终获得的分割效率相对HS和GHS更高。通过彩色图像的几种色彩空间(主要是RGB、HSV和HSI)的比较,将LSHS算法分别用于不同的色彩空间进行图像分割。实验效果显示LSHS对于各种彩色空间均能高效的分割,且能很好地改善HS陷入局部最优值的不足,并比HS、GHS拥有着更佳的稳定性和鲁棒性。
[Abstract]:With the rapid development of science and technology, the application of digital image in various fields is increasing day by day. Image segmentation is a basic technology of image processing, so people pay more and more attention to it. The method of image segmentation can be interpreted as dividing the image into parts with different features, then extracting the effective parts, and it is an important way for image processing to transition to analysis, and it has a very important position in the field of image. At the same time, it has been widely used in most fields and achieved good results. Nowadays, more and more intelligent algorithms are applied in the field of image segmentation. Most of the excellent intelligent algorithms have gradually replaced the usual methods, and are now the effective way to solve most optimization problems. Now the common intelligent algorithms are due to some of its own excellent characteristics with a small amount of computing time to achieve higher results. In recent years, Geem et al. have proposed a meta-heuristic algorithm, the harmony search (Harmony Search,HS) algorithm, which is compared with genetic algorithm, simulated annealing algorithm and Tabu search. Experimental results show that the performance of HS algorithm is good. However, there are still many problems to be solved in the research and application of HS. This paper mainly studies the improvement of the intelligent algorithm and uses the improved intelligent algorithm to improve the efficiency of the traditional segmentation algorithm. The work is as follows: firstly, several classical methods of image segmentation and the source, basic principle, concrete steps of HS are described, and their advantages and disadvantages are analyzed. This paper summarizes the problems faced by HS in engineering application and the main research directions at present, and discusses several classical improved HS.. Then, aiming at the deficiency that HS is easy to fall into local optimum, which leads to early convergence, a harmonic search (Local Search technique fusion of Harmony Search,LSHS algorithm combining local search is proposed in this paper. In the LSHS algorithm, the optimal harmonic vector and the two random harmonic vectors in the population are linearly combined to generate a new harmony, which expands the local search area and improves the convergence speed of the algorithm. The test results of LSHS and HS,GHS (Global-best Harmony Search,GHS) are compared with 9 standard test functions. The results show that the results of LSHS are better and the performance is better. Finally, because the optimization method can be used to find the optimal threshold, the LSHS proposed in this paper is applied to the maximum entropy segmentation. Three algorithms, HS,GHS and LSHS, are used to segment gray and color images, and the simulation results show that the segmentation efficiency of LSHS is higher than that of HS and GHS. By comparing several color spaces of color images (mainly RGB,HSV and HSI), the LSHS algorithm is applied to different color spaces for image segmentation. The experimental results show that LSHS can segment all kinds of color space efficiently, improve the deficiency of HS falling into local optimal value, and have better stability and robustness than HS,GHS.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】