基于颜色和边缘信息分布的图像检索

Image Retrieval Based on Color and Edge Distribution

提出了一种基于颜色和形状的图像检索方法。该方法首先将图像非等间隔量化到HSV颜色空间,在用SOBEL算子提取图像的梯度图(高频信息)后,根据HSV颜色聚合块,对梯度图进行必要的修剪,最后用不变矩来描述图像的形状特征;而颜色信息则用HSV颜色空间下的颜色直方图来表示。实验结果表明,该算法有较高的查准率,相比于基本颜色块的半径和角度直方图和普通颜色量化直方图,平均查准率分别提高了12.5%和6.6%。

A new retrieval method based on global color and shape (edge) features is proposed. This method modifies the gradient image generated by Sobel operator according the HSV Color Coherence Block, then describe shape feature using the Invariant Moments. And besides, Color information is described by HSV color quantification histogram in HSV color model. Results show that the average rates of precision of this method compared with RAHCB and common color quantification histogram is increased by 12.5% and 6.6% respectively.