摘要
基于内容的图像检索准确性大大依赖于低层可视特征的描述。本文提出一类创新的彩色图像空间描述子、纹理描述子、边缘描述子和颜色描述子,空间描述子由局部均值立方图表示,纹理描述子由局部方向差单元直方图表示,边缘描述子由局部极大-极小差直方图表示,颜色描述子由量化HSV模型颜色直方图表示。这四种描述子被用作持征索引,它们对彩色图像,尤其是对具有相对规则的结构或纹理特征的图像具有很强的描述力。实验结果表明,用这种特征索引来检索图像,可以得利比其它基于颜色-空间方法高得多的精确度。
The accuracy of content-based image retrieval depends greatly on the description of low-level visual fea- tures. In this paper,a novel spatial,texture,edge,and color descriptor for color image are proposed. Spatial descriptor is represented by local mean histogram,texture descriptor is represented by local directional difference unit histogram. edge descriptor is represented by local max-min difference histogram,color descriptor is represented by color his- togram under the quantized HSV model. These four descriptors are used as feature indexes,which have powerful de- scriptive power for the color images,especially for images with relatively regular structure characteristic or texture characteristic. Experimental results show that the retrieval method using the feature indexes can achieve much higher precision than other color-spatial based methods.
出处
《计算机科学》
CSCD
北大核心
2005年第2期216-218,F004,共4页
Computer Science
关键词
表示
彩色图像
图表
颜色直方图
纹理特征
索引
基于内容的图像检索
描述子
量化
边缘
Content based image retrieval (CBIR)
Color histogram
Local mean histogram
Local directional difference unit histogram
Local max-min difference histogram
