基于颜色分布特征的图像检索

Image Retrieval Based on Color Distribution Features

颜色直方图是基于内容的图像检索(CBIR)中的一种重要特征,然而其完全丢弃了图像颜色的空间分布信息,为了有效地利用图像颜色的空间分布信息,提出采用颜色分布熵(CDE)描述图像颜色的分布特征,并根据人的视觉特性及信息熵的特性,提出了进一步的改进算法。同以往的方法进行比较结果表明,该方法在图像的相似性检索时是很有效的,并具有较高的检索效率。

Color histogram is an important technique for color image database indexing and retrieving. However,the main problem of color histogram indexing is that it does not take the spatial relationship of the colors into consideration. Previous researches show that the effectiveness of image retrieval increases when the spatial relationship of colors is considered. We use the color distribution entropy （ODE） and improved ODE （I-ODE） to describe the global spatial relationship of colors. In comparison with the spatial chromatic histogram （SOH） and geostat,which also measure the global spatial relationship of colors, the experiment results show that ODE and I-ODE give better performance than SOH and geostat.