基于稀疏结构的图像特征匹配算法  

Image Feature Matching Algorithm Based on Sparse Structure

作　　者：包晓安[1] 詹秀娟 张俊为 王强[1] 胡玲玲 桂江生[1] BAO Xiao-An, ZHAN Xiu-Juan, ZHANG Jun-Wei, WANG Qiang, HU Ling-Ling, GUI Jiang-Sheng(School of Informatics and Electronics, Zhejiang Sci-Tech University, Hangzhou 310018, Chin)

机构地区：[1]浙江理工大学信息学院,杭州310018

出　　处：《计算机系统应用》2018年第4期178-183,共6页Computer Systems & Applications

基　　金：国家自然科学基金(61379036;61502430);国家自然科学基金委中丹合作项目(61361136002);浙江省重大科技专项重点工业项目(2014C01047);浙江理工大学521人才培养计划(20150428)

摘　　要：对图像进行全局的特征点检测耗时较长,而且全局特征稳定性不好,这就造成算法的运行速度慢和匹配准确率低,达不到令人满意的匹配效果.在尺度不变特征变换(SIFT)的基础上,通过引入稀疏结构的概念,提出了一种基于稀疏结构的图像特征匹配算法(SSM).通过稀疏度函数获得像素点的稀疏度值,筛选出稀疏度高的像素点所在的区域,并对该区域进行SIFT特征点检测,通过最佳描述子实现特征匹配.将SSM算法与几种经典算法相比,实验结果表明,本文算法在特征匹配速度和匹配准确率上相比于原算法都有较明显的提高,能够用于目标实时跟踪、图像检索和全景图像拼接等领域.The global image detection of feature points is time-consuming, and the global feature is not of good of stability, which causes the algorithm speed to be slow and the matching accuracy to be low, with the matching effect satisfactory. On the basis of scale invariant feature transform （SIFT） based on the sparse structure of the concept, this study puts forward an image feature matching algorithm based on sparse structure （SSM）. It gets the pixel value by sparse sparse degree function, selects pixel highly sparse region, and detects the SIFT feature point of the region, to achieve feature matching by using the best descriptors. Compared with several classical algorithms, the experimental results show that this algorithm has significantly improved in feature matching speed and accuracy, and it can be used for real-time object tracking, image retrieval and image mosaics, and other fields.

关 键 词：稀疏结构 相似度 描述子 直方图 特征匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]