摘要
图像分割是模式识别和图像分析的预处理阶段,通常采用聚类的方法进行.图像分割技术被认为是计算机视觉中的一个瓶颈.基于扩展的Otsu最优阈值图像分割方法,提出了一种用遗传算法进行图像分割的方法,并给出了遗传算法中基本参数的设定.实验结果表明,基于图像的像素方差信息,利用遗传算法全局搜索图像的双阈值,这样不但图像分割效果好,而且缩短了计算时间,并具有遗传算法鲁棒性和自适应的特点,比传统的Otsu方法有明显的优点.在遗传算法中引入了优生算子、变异算子和新个体,避免了局部早熟,提高了收敛速度和全局收敛能力.GA作为一种并行算法,提高速度的潜力十分巨大.
Image segmentation, as the pretreatment of the pattern recognition and image analysis, is processed by clustering. The technique of image segmentation is regarded as the bottleneck of the computer vision. An image segmentation method by genetic algorithms （GA） is proposed, which is based on optimal Otsu threshold algorithm. The basic parameters of GA are defined for image segmentation. Based on the information of square errors of image pixels, GA is used to search double thresholds. The experimental results indicate that it is not only of higher segmentation quality, but also reduces the running time and is of robustness and self-adaptabihty, and that it is better than traditional Otsu method. The paper introduces prepotency and variance operator and new individuals, so the arithmetic avoids premature and improves convergent speed and capability. GA, as a kind of parallel computing, is huge in the potential to improve its computing speed.
出处
《南京师范大学学报（工程技术版）》
CAS
2007年第1期14-17,36,共5页
Journal of Nanjing Normal University(Engineering and Technology Edition)
关键词
阈值
图像分割
遗传算法
threshold, image segmentation, genetic algorithm
