一种新的用于图像检索的颜色——空间特征  

A Color-Spatial Feature for Content Based Image Retrieval

作　　者：木拉提.哈米提 刘伟[2] 孙静[1] 姬金虎[1] 艾克热木.阿西木 毛莉萍[1] 胡彦婷[1] 

机构地区：[1]新疆医科大学医学工程技术学院,乌鲁木齐830054 [2]杭州电子科技大学自动化学院生物医学工程及仪器研究所,杭州310018

出　　处：《科技通报》2008年第5期696-699,共4页Bulletin of Science and Technology

基　　金：新疆维吾尔自治区少数民族科技人才特殊培养科研项目专项基金(200723104)

摘　　要：为了解决图像检索中颜色直方图无法描述颜色空间位置信息的问题,提出了一种新的基于色彩量化的凸壳内角直方图特征。针对量化后HSV颜色空间中某一具体量化颜色值对应像素点的空间分布,首先计算包含这些像素点位置信息的凸壳,并对该凸壳的所有内角建立直方图。然后处理所有量化颜色值提取整幅图像的凸壳内角直方图特征。将该方法和一些文献中提出的方法进行的比较实验结果表明:提出的新特征可以有效地描述颜色空间位置信息,在查准率和查全率指标上均优于已有文献中的一些方法。In order to solve the problem that color histogram can not describe the color spatial information of an image,a convex-hull intra-angle histogram feature based on color quantization was proposed. Firstly ,the image was quantized in HSV color space. Secondly,for a given color quantization value in HSV color space,the convex-hull that includes all pixels whose color quantization values are equal to the given value was computed. Then a histogram was computed based on the all intra-angle （s）of the convex-hull. Finally,the convex-hull intra-angle histogram of the whole image was built by processing all color quantization values. Comparision study was carried out between other similar approaches proposed in literatures and ours. Experimental results showed that our approach can effectively describe the color spatial information. Furthermore, better retrieval accuracy can be achieved by using our approach.

关 键 词：图像检索 颜色-空间特征 直方图 凸壳 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP751.1[自动化与计算机技术—计算机科学与技术]