基于空间结构统计建模的图像分类方法  

Statistical modelling of spatial structures based image classification

作　　者：刘金平[1] 唐朝晖[2] 朱建勇[2] 谭珍[2] 

机构地区：[1]湖南师范大学数学与计算机科学学院,长沙410081 [2]中南大学信息科学与工程学院,长沙410083

出　　处：《控制与决策》2015年第6期1092-1098,共7页Control and Decision

基　　金：国家自然科学基金项目(61472134;61171192;61272337)

摘　　要：提出一种基于图像空间结构统计建模的复杂纹理图像模式识别方法.从理论上分析了复杂纹理图像空间结构的韦伯分布过程,通过构造多尺度全向高斯导数滤波器,获得复杂纹理图像在不同观测尺度上的全方向空间结构统计建模表征结果.基于偏最小二乘-判决分析原理构建分类器,实现了复杂纹理图像的分类识别.实验结果表明,所提出的图像空间结构统计建模方法能获得复杂纹理图像关健性的视觉感知特性,基于该方法的图像分类准确率高且性能稳定.A method of statistical modeling of spatial structures based complex textural image pattern recognition is presented. The Weibull distribution process of the spatial structures in the complex textural images is analyzed theoretically in advance. Subsequently, by filtering the images with a proposed filter bank of Gaussian derivative filters of multiscale and omnidirectional, the multi-scale omnidirectional spatial structure features of the complex textural images are obtained by statistical modeling of the image’s spatial structures. Based on the principle of partial least squares discriminant analysis, the image texture classification recognition model is established, which can be used to identify the pattern of the complex texture images effectively. Experimental results show that, the proposed image statistical modeling method achieves distinctive visual perception characteristics of the complex textural images, which has strong classification ability and stable classification performance.

关 键 词：连续分裂理论 统计建模 韦伯分布 偏最小二乘判决分析 复杂纹理图像分类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]