基于数据共性的鲁棒特征表示  

Robust feature representation based on data commonalities

作　　者：岳金朋[1,2] 史忠植[1] 张伟[2,3] 胡宏[1] 张建华[1,2] 

机构地区：[1]中国科学院计算技术研究所智能信息处理重点实验室,北京100190 [2]中国科学院大学,北京100049 [3]中国科学院计算技术研究所前瞻研究实验室,北京100190

出　　处：《高技术通讯》2014年第2期117-123,共7页Chinese High Technology Letters

基　　金：973计划(2013CB329502);国家自然科学基金(61035003,60933004);863计划(2012AA011003)资助项目

摘　　要：为了提高图像检索的性能,研究了图像变化对视觉特征鲁棒性的影响,提出了一种新的鲁棒特征表示方法。该方法首先挖掘鲁棒特征的数据共性,即在不同图像变化条件下提取的鲁棒特征之间的共性,然后基于数据共性进行特征的二进制码表示。在特征挖掘阶段,根据特征的视觉信息和在向量空间下的相似性来挖掘数据共性。在特征表示阶段,对具有共性的特征进行离线学习,通过局部保持哈希(LPH)将具有共性的数据表示为相似的二进制码。该方法由于将特征提取过程中的潜在信息即数据共性与特征表示技术相融合,因此能够更好地应对复杂图像变化。实验表明,在图像检索应用背景下该方法的精度比现有方法提高6%以上。To improve the performance of image retrieval, the effects of image variations on the robustness of visual fea- tures were studied, and then a novel robust feature representation method was proposed. The method mines robust features＇ data commonalities, the general characters among the robust features extracted under different image varia- tions, and then represents features in binary codes based on the data commonalities. In the feature mining stage, the commonalities are obtained based on the visual information and the similarities of features in a vector space. In the feature representation stage, the commonalities of features are off-line learned and the data with commonalities are represented as similar binary codes by using the locality preserving Hash （LPH）. This method can perform well un- der complex image variations, for it fuses data commonalities implicit in the feature extracting process with feature representation techniques. The experimental results show that the precision of the method can be improved by more than 6% compared with the existing methods in image retrieval applications.

关 键 词：图像内容分析 视觉特征提取 鲁棒特征表示 数据共性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]