基于遗传算法的图像分割研究

【作者】 胡硕；

【导师】 乔双；

【作者基本信息】 东北师范大学 ， 电路与系统， 2003， 硕士

【摘要】 遗传算法具有简单、鲁棒性好和本质并行的突出优点。其在应用领域取得的巨大成功，引起了广大学者的关注。在图像分割领域，遗传算法常用来帮助确定分割阈值。 本文讨论了目前遗传算法应用于图像分割的现状，给出了几种遗传分割算法的原理、过程、实验结果及分析； 介绍了图像边缘检测、图像阈值分割的各种算法，并给出了对比分析；对遗传算法的基本概念和研究进展进行了综述，提出了一种新的遗传分割算法，得到了理想的结果。 本文提出的遗传分割算法充分考虑了图像数据本身的特殊性，从提高全局搜索能力和收敛速度出发，加入了3个新的操作策略。算法在初始化种群阶段引入了“优生”算子，以及改进的变异操作使算法的收敛速度大大提高；在形成新种群阶段引入新的算子避免了局部早熟，提高了全局收敛能力。本文以基于坐标的阈值分割方法为基础进行二维整数编码，采用窗口交叉方法，以文献[23]给出的评价方法构造适应度函数。实验结果表明，本文提出的遗传分割算法明显优于传统分割算法。 本文所有程序均是用VC++6．0在Win98环境下编译完成。实验图片源于实际拍摄的图片及网上收集的图片。更多还原

【Abstract】 Genetic algorithm (GA) has the virtue of simpleness, robustness and parallel in essence. It has been applied perfectly in the engineering field, which appeals to many scholars in the world. In the image segmentation field, GA is usually used to get the threshold of image segmentation.The status of GA applied in the image segmentation field recently is presented, and the theories, steps, results and analyses of several GAs applied in the image segmentation are given.Algorithms and analyses about edge detection and threshold selection of the image segmentation are presented. An overview of the basic theories and the recent development is given, and a new genetic algorithm applied in image segmentation (GAS) is presented.Considering image data is often very massive, GAS introduces three new measures in order to solve the problem of global convergence and improves the convergence speed. Introduction of prepotency operator in the initialize population step and the improved mutation operator accelerate the convergence process, and the introduction of new operator in forming new population step avoid converging in local optimum, and promote the ability of global convergence. Coding based upon image threshold segmentation related with coordinates, using windows crossover method, designing evaluation function based upon the equations given in literature [23], GAS gets much better results than traditional algorithm.Programs were all compiled in the Win98 by VC++6.0. All photos were collected from Internet and personal photos.更多还原