一种基于分形理论的多尺度多方向纹理特征提取方法  

Multi-scale and multi-orientation texture feature extraction method based on fractal theory

作　　者：赵莹[1] 高隽[1] 陈果[1] 冯文刚[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《仪器仪表学报》2008年第4期787-791,共5页Chinese Journal of Scientific Instrument

基　　金：新世纪优秀人才支持计划(NCET-04-0560)资助项目

摘　　要：纹理分割是计算机视觉以及图像处理中一项重要的任务。基于分形理论的纹理分割方法是一种常用的方法,然而单一采用分形维数作为特征不足以描述纹理。针对这一问题,本文提出了一种基于分形理论的多尺度多方向纹理特征提取方法。有效地提取出具有尺度性、方向性的分形特征,是对纹理图像的一种多层次的、紧致的描述。文中定义了一种新的相似度测量方法,采取基于图论的归一化切分方法作为分割策略。通过实验,验证了该方法的有效性和通用性。In machine vision and image processing, texture segmentation is a very important task. The texture segmentation based on fractal geometry is a commonly used method. However, it is inadequate only using fractal dimension to describe the texture. In this paper, a multi-scale and multi-orientation method of texture feature extraction based on fractal theory is proposed. The method extracts fractal features with scale and orientation characteristics, which are multi-scale and tightened description for the texture. A new similarity measurement method is defined. Normalized cut method is adopted as the segmentation algorithm. Simulation results of texture segmentation experiments prove the effectiveness of the proposed method.

关 键 词：纹理分割 分形维数 方向可调金字塔 归一化切分 人类视觉特性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]