基于广义平均直方图的图像检索方法

Generalized Average Histogram Based Image Retrieval

由于传统直方图方法不能描述图像中相邻像素之间的相关性和颜色在图像内的空间分布特征,文中引入"广义图像"的概念,即把原图像与平滑图像相对应的两个像素组合起来,形成的一个二元组。提出了广义平均直方图的概念,首先将颜色量化方法扩展到广义图像中,把广义图像的颜色量化为143维,然后计算每一维对应的聚类中心和组合数,最后计算每一维对应的频数,得出广义平均直方图。为了提高对多色彩图像的检索效率,把自动参考颜色表方法引入图像匹配中。实验结果表明,基于广义平均直方图的图像检索方法的有效性好、准确性高。

Conventional color histogram approaches to image retrieval lack the expressions of spatial information and relativity between adjacent pixels. In this paper,we introduce ＂Generalized Image＂ （GI）,which combines the original image with its smoothed image to form a binary set,to mitigate this problem. And we present a new retrieval method based on generalized average color histogram. In the algorithm,color quantifier is extended to GI,a feature vector of 143 dimensions is get successfully also. After computing the cluster centers for each dimension and counting the number of combination number,an improvement on the conventional color histogram called generalized average color histogram （GACI） is proposed. Moreover, for better muhicolor matching,the 143 dimensions are arranged according to pixel numbers of each bin, then index images by the references color set algorithm. Compared with the typical histogram method ,our new method in general achieves more accurate and efficiency.