用于纹理检索的多尺度复杂性纹理描述子  

Multi-scale complexity texture descriptors for texture retrieval

作　　者：刘伟 徐伟栋[1] 厉力华[1] 

机构地区：[1]杭州电子科技大学生命信息与仪器工程学院,浙江杭州310018

出　　处：《电路与系统学报》2011年第1期25-30,共6页Journal of Circuits and Systems

基　　金：浙江省自然科学基金(Y1080740);国家杰出青年科学基金(60788101);国家自然科学基金(61001215;60705016);浙江省重大科技攻关国际合作项目(2006C14026)

摘　　要：提出了基于复杂性度量和图像多尺度分析的纹理特征提取方法,提取后的特征用于纹理检索。使用的复杂性度量为一维和二维C0复杂性。采用的图像多尺度分析方法有金字塔分解、小波分解(Haar小波与Db2小波)和二维经验模式分解。实验结果表明:提出的基于Db2小波分解的一维和二维C0复杂性特征是适于纹理检索的特征。它们均取得了和Gabor特征相近的检索结果,计算特征所需要的时间比Gabor特征少一个数量级。In this paper, a study of two complexity measurements and multi-scale image analysis based texture descriptors is reported and applied for texture retrieval. A time-series complexity measurement and a two-dimensional data complexity measurement, which are called CO complexity and 2D-C0 complexity respectively, are compared for texture feature extraction and retrieval. Three multi-scale image analysis approaches including pyramid decomposition, wavelet decomposition （Haar and Db2 wavelet） and bidimensional empirical mode decomposition（BEMD） are also investigated. Experimental results show that the proposed Db2-wavelet based features achieve fairly good retrieval accuracy for texture retrieval. They are computationally more efficient than that based on Gabor feature, whereas obtaining comparable texture retrieval accuracy.

关 键 词：图像检索 纹理分析 特征提取 复杂性测度 多尺度分析 

分 类 号：O231.5[理学—运筹学与控制论] TP391.4[理学—数学]