摘要
特征提取是图像检测或图像检索的关键步骤,SIFT特征能够实现平移、旋转、缩放等不变性,MSER特征实现了仿射不变性。集成SIFT和MSER特征的优势,提出了一种图像的融合特征提取方法,融合特征相比单一的局部特征具有更好的鲁棒性,还实现了图像特征的加速匹配,同时融合特征减少了存贮空间。针对这种图像的融合特征表示方法,给出了相应的图像匹配策略,实验结果表明提出的融合特征及检测方法在INRIA copy dataset数据集上取得了很好的效果。
Extracting feature is the key of duplicate image detection. SIFT shows the invariant of shift, rotation and scale and MSER shows the invariant to affine wansformation. Image fusional feature representation is presented with SIFT and MSER in this paper, which integrates SlFr and MSER. The funsional feature is robuster than a single local feature. At the same time, the speed of match is acceler- ated by using the fusional feature. The fusional feature needs lower storage space than single SIFT feature. The image match strategy is designed corresponding to fusional feature representation. The experimental results show the effective performance of the proposed meth- od on INRIA copy image dataset.
出处
《计算机技术与发展》
2012年第8期103-106,共4页
Computer Technology and Development
基金
辽宁省教育重点实验室项目(LS2010079)
