基于基元相关性描述子的图像检索  

Image Retrieval Based on Texton Correlation Descriptor

作　　者：吴俊[1] 刘胜蓝[2] 冯林[1,3] 于来行[3] Wu Jun;Feng Lin;Liu Shenglan;Yu Laihang(School of Innovation and Entrepreneurship,Dalian University of Technology,Dalian,Liaoning 116024;School of Control Science and Engineering,Faculty of Electronic Information and Electrical Engineering,Dalian University of Technology,Dalian,Liaoning 116024;School of Computer Science and Technology,Faculty of Electronic Information and Electrical Engineering,Dalian University of Technology,Dalian,Liaoning 116024)

机构地区：[1]大连理工大学创新创业学院,辽宁大连116024 [2]大连理工大学电子信息与电气工程学部控制科学与工程学院,辽宁大连116024 [3]大连理工大学电子信息与电气工程学部计算机科学与技术学院,辽宁大连116024

出　　处：《计算机研究与发展》2016年第12期2824-2835,共12页Journal of Computer Research and Development

基　　金：国家自然科学基金项目(61173163,61370200);中国博士后科学基金项目(ZX20150629)~~

摘　　要：图像检索系统性能很大程度上取决于提取的图像描述子,其中颜色差分直方图(color difference histogram,CDH)已经在图像检索中显示出了较好的性能.但是这种描述子仍然有一定的局限性:1)只考虑到了像素间颜色差分的整体分布;2)忽略像素间的空间位置分布.因此提出了1种新的基元相关性描述子(texton correlation descriptor,TCD)提取图像特征,并将其应用于图像检索系统中.具体提取过程分为3个步骤:1)利用图像底层特征(颜色和局部二值模式)检测一致性区域,选择图像中包含区分性信息的局部区域;2)提出颜色差分特征和基元频率特征分别描述图像像素间的对比度和空间位置信息,其中颜色差分特征融合了描述局部邻域的颜色差分相关性统计和全局颜色差分直方图,基元频率特征也融合了描述局部邻域的基元频率相关性和基元频率直方图;3)联合一致性区域中的这2种特征得到最后的TCD描述子.这种特征描述了图像中2种互相独立并互相补充的特性:对比度和空间位置关系,并同时考虑到了这2种特性在局部和全局区域中的描述,因此在图像检索实验中会有更好的性能.在图像数据集中的实验结果显示了TCD描述子的检索效果明显优于其他几种特征描述子,证实了TCD描述子在图像检索中的有效性和稳定性.The performance of content-based image retrieval(CBIR)depends to a great extent on theimage feature descriptor.Among these descriptors?color difference histogram(CDH)has showed thegreat discriminative performance in CBIR.However,there are still some limitations in it:1)onlytaking color difference of pixels in global region into account;2)not considering the spatial structureamong pixels.In this paper,to solve these problems,we propose a novel image representation,calledtexton correlation descriptor(TCD),which is applied to CBIR.F irst,we define uniform regionswhich contain discriminative information of images and then detect them by analyzing the relationshipamong low-level features(color value and local binary patterns)of pixels.Second,in order tocharacter contrast and spatial structure information in uniform regions respectively,we propose thecolor difference feature which fuses color difference correlation and global color difference histogram,and texton frequency feature which fuses texton frequency correlation and texton frequencyhistogram.Finally,by combining these feature vectors,TCD not only characters two orthogonalproperties:spatial structure and contrast,but also takes these properties in local and global uniformregions into account simultaneously so that TCD has better performance in CBIR.The experimentalresults show that the retrieval results of TCD is higher than that of other descriptors in imagedatasets,and thus demonstrate that TCD is more robust and discriminative in CBIR.

关 键 词：基元相关性描述子 一致性区域 颜色差分特征 基元频率特征 图像检索 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]