一种基于Curvelet相关图纹理图像检索方法  

A texture image retrieval approach based on Curvelet correlogram

作　　者：李雷[1] 

机构地区：[1]西南财经大学西部商学院,四川成都610074

出　　处：《电子技术应用》2016年第7期119-121,129,共4页Application of Electronic Technique

基　　金：四川省科技厅支撑项目(2012GZX0080)

摘　　要：纹理是图像普遍存在且难以描述的特征,但是更容易引起人们视觉上的关注,它蕴含有丰富的信息,所以基于纹理的检索具有重要的应用价值。提出了一种新的基于Curvelet相关图纹理图像检索方法。该方法通过对纹理图像进行Curvelet分解进行多尺度分析;利用变换后粗尺度反映图像轮廓、细尺度反映图像纹理信息特性,阈值处理后采用不同的量化等级;计算Curvelet系数相关图,反映了不同系数所占比例和相互之间的空间相关性,由此构造图像的特征向量。通过对Brodatz纹理图像库实验,结果表明该方法相较于原有的Curvelet方法能够更有效地进行纹理图像检索。Texture is a ubiquitous feature of image and is easily to be perceived,whereas it is difficult to be described.Texture features of most images contain a wealth of information and texture offers a new clue to image retrieval.In this paper,a new algo-rithm is presented for texture image retrieval.According to the new algorithm,firstly the Curvelet transform is applied for texture image multi-scale decomposition.Since the coarse scale reflects the contour of image and the fine ones reflect the texture informa-tion,different quantization levels are used after threshold processing,and then auto-correlograms of Curvelet coefficients are comput-ed which show the proportion and spatial correlation of different coefficients.Finally,feature vectors are constructed for similarity comparison.Experiments based on Brodatz benchmark database turn out that the new algorithm has an improvement in effectiveness compared with other retrieval methods based on Curvelet in texture image retrieval.By using the new method,image texture fea-tures can be expressed better,which leads to the more accurate retrieval.

关 键 词：多尺度分析 Curvelet相关图 纹理图像检索 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]