基于多小波和非下采样Contourlet变换的纹理检索  

Image Retrieval Based on Multi-wavelet and Non-subsampled Contourlet Transform

作　　者：党留群[1] 张慧云[1] 张新明[1] 李双[1] 牛振齐[1] 

机构地区：[1]河南师范大学计算机与信息技术学院,河南新乡453007

出　　处：《河南师范大学学报（自然科学版）》2011年第3期153-156,共4页Journal of Henan Normal University(Natural Science Edition)

基　　金：河南省重点科技攻关项目(092102210017);河南省教育厅科技攻关项目(2007520024;2008B520021)

摘　　要：为了更有效地进行纹理图像检索,提出了一种新的基于多小波变换和非下采样Contourlet变换相结合的纹理图像检索方法.首先,采用多小波对图像进行多尺度分解以便保证变换平移不变性的同时也消除变换的冗余性;然后,将所得的高频子带采用非下采样方向滤波器组进行多方向分解,计算低频子带和各层方向子带的一阶几何矩作为纹理特征;最后用欧式距离进行相似性度量.实验表明,该方法与多分辨理论检索的传统方法相比具有较高的查全率和查准率.To improve the quality of texture image retrieval,this paper proposes a new method for texture image retrieval based on multi-wavelet and non-subsampled contourlet transform.Firstly,the image is decomposed into multi-scale with multi-wavelet to ensure translation invariant and eliminate transform redundancy.Then,multi-directional decomposition of the high-frequency sub-bands is implemented with non-subsampled directional filter bands in every scale,the first-order geometric distances of the low frequency sub-band and every scale's directional sub-band are calculated as the texture features.Finally,the retrieval results are obtained using the Euclidean distance.The experimental results show that the proposed method's retrieval performance is better than that of the traditional multi-resolution approaches.

关 键 词：纹理图像检索 多小波 非下采样CONTOURLET变换 几何矩 

分 类 号：N55[自然科学总论] G658.3[文化科学—教育学]