基于DCT系数和颜色直方图的图像检索  

Image Retrieval Based on DCT Coefficients and Color Histogram

作　　者：谢莉[1] 余胜[1] 成运[1] 

机构地区：[1]湖南人文科技学院信息科学与工程系,湖南娄底417000

出　　处：《邵阳学院学报（自然科学版）》2015年第4期16-23,共8页Journal of Shaoyang University：Natural Science Edition

基　　金：国家自然科学基金(61165011);湖南省科技计划项目(2013GK3109);湖南省自然科学基金项目(12JJ2040);湖南省重点学科建设项目;湖南省高校科技创新团队支持计划项目;湖南省教育厅资助项目(15C0726);湖南人文科技学院教学改革研究项目(RKJGZ1419)

摘　　要：为有效描述图像的形状、纹理和颜色特征,提出一种基于DCT系数统计特征和颜色直方图的图像检索.首先,将图像中像素点颜色值在RGB颜色空间中量化到64色颜色空间,构建彩色图像的颜色直方图用于描述图像的颜色特征.然后,将图像划分为8×8子块,分别对各子块进行DCT变换.最后,根据各子块的DCT系数获得其两个统计特征,其一是计算各子块DCT系数的均值和方差,构建均值-方差直方图用于描述图像的纹理特征,另一个是根据各子块的前9个AC系数构建AC系数差值直方图用于描述图像的局部形状特征.结合颜色直方图、均值-方差直方图和AC系数差值直方图构成图像检索的特征向量.实验结果表明,算法有效描述了图像的形状、纹理和颜色特征,查全率和查准率均有提高,检索性能较优.In order to describe shape,texture and color features,this paper presents an image retrieval algorithm based on DCT coefficients statistical characteristics and color histogram. Firstly,quantize the pixel values into 64 colors,and construct color histogram to describe color feature. Secondly,divide the image into 8×8 sub-block,perform DCT on each sub-block. Lastly,according to the DCT coefficients of each block to obtain the two statistical characteristics,one is to calculate the mean and variance of each sub block DCT coefficient,and construct the mean-variance histogram to describe texture feature; the other is to construct AC coefficient difference histogram by using those front 9 AC coefficients so as to describe the local shape feature. A new image retrieval feature vector is formed by combining the color histogram with the mean-variance histogram and the AC coefficient difference histogram.Experimental results indicate that the algorithm can describe the shape,texture and color features and has higher precision and recall rate,in all,it has a better performance than other image retrieval methods.

关 键 词：图像检索 离散余弦变换 离散余弦变换系数 颜色直方图 均值-方差直方图 AC系数差值直方图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]