基于双广义高斯模型和多尺度融合的纹理图像检索方法  

Texture Image Retrieval Method Based on Dual-generalized Gaussian Model and Multi-scale Fusion

作　　者：杨娟[1] 李永福[1] 汪荣贵[1] 薛丽霞[1] 张清杨 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《电子与信息学报》2016年第11期2856-2863,共8页Journal of Electronics & Information Technology

基　　金：中国博士后基金(2014M561817);安徽省自然科学基金(J2014AKZR0055)

摘　　要：纹理因素是描述图像的重要特征之一,为了准确地刻画纹理特征,增强图像的区分能力,该文提出一种基于双树复数小波域统计特征的纹理图像检索方法。首先对图像采用双树复数小波变换得到各子带系数,由于系数存在细微不完全对称分布特性,将其建模为双广义高斯模型。其次,因为各子带系数之间不完全独立也不完全冲突,存在不确定关系,所以采用模糊集合和证据理论(FS-DS)的方法,融合各子带系数特征。最后,对Brodatz和彩色纹理图像库进行仿真实验,并与多种统计建模的方法相比较。结果表明,该方法有效地提高了纹理图像的平均检索率。Texture factor is one of the most important characteristics in the image description. In order to describe the texture feature accurately, and enhance image distinguish ability, a method of texture image retrieval is proposed based on Dual-Tree Complex Wavelet Transform （DT-CWT） in this paper. Firstly, each sub-band coeffmient is obtained by DT-CWT, because the coefficient distribution exists slight incomplete symmetrical feature, which is modeled as dual-generalized Gaussian model. Secondly, there is incomplete independent and uncertain conflict between the sub-band coefficients, therefore the Fuzzy Set and Dempster-Shafer （FS-DS） evidence theory are applied to blending the characteristics of each subband coefficients. The performance of the propose algorithm is tested on the Brodatz and color texture image library, and also compared with a variety of statistical modeling methods. The experimental results demonstrate that the proposed method can improve the average retrieval rate of the texture images effectively.

关 键 词：纹理图像检索 双树复数小波 双广义高斯分布 模糊集合 证据理论 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]