摘要
提出一种创新的基于多邻域统计矩直方图方法(MNSMH),该方法在量化HSV颜色模型下,计算每个像素点不同邻域的统计矩,对每个邻域统计矩,计算它的归一化直方图,以这些直方图和颜色直方图一起作为图像的特征索引进行彩色图像检索.这些不同邻域矩反映了图像颜色的空间分布信息,而它们的直方图又是对整个图像的全局统计,对图像的平移、旋转和尺度不变.实验结果表明,该方法性能稳定,与两种基于颜色直方图方法相比,能够明显地提高检索率.
A novel color image retrieval method based on multi-neighbor statistic moment histograms (MNSMH) is proposed. In MNSMH, the statistic moments of different neighbor of a pixel are calculated under the quantized HSV color model. For every neighbor statistic moment, its normalized histogram is obtained. These histograms and color histogram are used as feature index to retrieve image. These neighbor statistic moments capture the spatial distribution information of the color, at the same time, their histograms are global statistic over whole image and are invariant to image translation, rotation and scaling. The experimental results show that the method has stable performance and provides higher retrieval precision than two color histogram-based methods.
出处
《小型微型计算机系统》
CSCD
北大核心
2005年第6期1061-1064,共4页
Journal of Chinese Computer Systems
关键词
基于内容的图像检索
颜色直方图
邻域统计矩直方图
content-based image retrieval(CBIR)
color histogram
neighbor statistic moment histogram
