文献类型：期刊文献

中文题名：基于基元自相关图和结构元直方图的图像检索

英文题名：Image retrieval based on texton autocorrelograms and structure element histogram

作者：刘芳辉[1];郭慧[1];张培[1];周邵萍[1]

机构：[1]华东理工大学机械与动力工程学院,上海200237

年份：2017

卷号：43

期号：8

起止页码：115

中文期刊名：电子技术应用

外文期刊名：Application of Electronic Technique

收录：CSTPCD;;北大核心:【北大核心2014】；

基金：国家自然科学基金(51575185)

语种：中文

中文关键词：图像检索;基元自相关图;结构元直方图;颜色代表值;相似度

外文关键词：image retrieval; texton autocorrelograms; structure element histogram; color represents values; similarity

摘要：针对目前的基元自相关图在表达图像颜色特征时空间相关性不强、基于单一特征检索时精度低的问题,提出了一种改进的基元自相关图和结构元直方图的图像检索方法。首先,将彩色图像非均匀的量化为63种颜色,并且对图像进行均匀分块,同时采用子块颜色代表值代替子块内每个像素点的颜色值;然后用定义的基元和结构元分别统计每个子块信息并提取图像基元自相关图和结构元直方图;最后综合提取的两个特征进行相似性度量。实验结果表明,相较于以颜色自相关图、基元自相关图及基于结构元的图像检索方法,该方法有效地提高了检索的精准率,改善了检索结果的排序值,并具有很好的稳定性。
In view of the present primitive texton autocorrelograms in expression image color feature spatial correlation is not strong,based on single feature retrieval problem of low accuracy,this paper proposes an improved retrieval method based on the texton autocorrelograms and structure element histogram image.Firstly,the color image is quantized to 63 colors nonuniformly,and the image is evenly partitioned,the sub-block color representative value is used to replace the color value of each pixel in the sub-block.Then,the information of each sub-block is statistically calculated by using the primitives and the structural element,and the texton autocorrelograms and structural element histogram of the image are extracted.Finally,the similarity measure is extracted from the two extracted features.The experimental results show that compared with color autocorrelograms and two other image retrieval methods,the algorithm effectively improves the precision of retrieval rate,it has improved the search results sorted value and have good stability.