提出一种基于稳定兴趣点空间分布的图像检索方法。首先使用基于优化梯度滤波(ODF)的兴趣点检测器,结合尺度归一化的方法,检测图像中的稳定兴趣点;再利用稳定兴趣点的空间分布,对图像进行环形和凸包区域划分;最后利用凸包颜色直方图和环形区域兴趣点邻域内伪泽尼克矩的加权特征向量,对图像进行特征描述。实验表明,本方法实现简单,对图像旋转、平移具有鲁棒性;与其他基于兴趣点的检索方法相比,降低了不稳定兴趣点的影响,平均检索速度较快且平均检索准确率提高了7.0-10.9%,可以更准确地查找到用户所需图像。
A new image retrieval method based on optimal derivative filter is presented. Firstly, the interest point detector based on optimal derivative filter is used to determine the location of stable interest points in the scale-normalized iraage. Then, the convex hull and annular color histogram in the local field of the stable interest points are calculated to depict the features. Experiment results show that the method is simple to realize,robust to rotation and translation. Compared with other interest-points-based retrieval methods,it can improve the average retrieval precision by 7.0%- 10. 9% ,and efficiently over- come the influence of unstable interest points.