摘要
针对传统基于全局特征的图像检索方法存在的不足,提出一种基于显著点特征和SVM(support vector machine)相关反馈相结合的图像检索方法.显著点提取方法是对图像进行小波分解,选择粗分辨率下绝对值较大的小波系数,它们对应原图像中变化较大的区域,然后在细分辨率下跟踪这些小波系数,提取原图像中的能代表这些变化的点,即显著点;然后利用显著点的空间分布信息,提取显著点周围局部区域的特征进行检索,并对检索结果进行SVM相关反馈.实验结果表明,引入反馈的方法可有效地检索更多的相关图像,明显提高了检索的准确性.
To solve the problem of the traditional image retrieval based on global features, a novel method for image retrieval based on salient points and SVM （ support vector machine ） relevance feedback is presented. Firstly, the images are decomposed using wavelet transform. We choose high wavelet coefficients in absolute value at a coarse resolution, because they correspond to a region with high global variations, then look at wavelet coefficients at finer resolutions to find the relevant points to represent the global variations. Then, the local feature of salient points is extracted for image retrieval, which utilizes the distribution information of salient points. ？ Finally, the retrieval results are resorted using SVM relevance feedback. Experimental results show that the feedback method can retrieve more relevant images, improve the retrieval accuracy significantly.
出处
《小型微型计算机系统》
CSCD
北大核心
2012年第1期173-177,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61071173)资助
关键词
图像检索
小波显著点
SVM相关反馈
image retrieval
wavalet-based salient points
SVM relevance feedback
