Blocking Contourlet Transform: An Improvement of Contourlet Transformand Its Application to Image Retrieval
Blocking Contourlet Transform: An Improvement of Contourlet Transformand Its Application to Image Retrieval
复制标题
DOI：
10.4304/jcp.7.9.2262-2268
复制
发表时间：
2012-01
期刊：
影响因子：
--
通讯作者：
Jian Wu;Zhiming Cui;Pengpeng Zhao;Jianming Chen
中科院分区：
文献类型：
--
作者：
Jian Wu;Zhiming Cui;Pengpeng Zhao;Jianming Chen
轮廓波变换是解决二维或多维奇异性的有效方案，具有良好的方向性和各向异性。针对描述物体边缘信息空间分布特征能力的不足，提出一种基于Contourlet变换的图像检索新算法，对索引图像进行分块，并利用Contourlet变换分解各个子块图像。首先，对每个子块图像的子带数据进行加权处理，从高低频子带数据中提取分类能力高的特征，对分类能力高的特征给予较大的权重。然后根据各个子块图像的能量，对纹理特征较强的子块图像给予较大的权重。最后，利用两个图像特征向量之间的加权欧氏距离作为图像相似度来检索图像。实验结果表明我们的算法具有良好的检索性能。
Contourlet transform is an effective solution to solve two or more dimensional singularity and has good direction and anisotropy. Against the shortage of ability of describing the spatial distribution characteristic of object’s edge information, this paper proposed a new image retrieval algorithm based on Contourlet transform, which blocks the indexed image and decomposes each sub-block images using Contourlet transform. At first, carry out weighted processing for sub-band data of each sub-block image, extract features with high classification ability from high and low frequency sub-band data, and give greater weight for those features with high classification ability. Then, according to the energy of each sub-block image, give greater weight for those sub-block image with strong texture characteristic. At last, retrieve the images using weighted Euclidean distance between two image feature vectors as image similarity. The experiment results show that our algorithm has good retrieval performance.