基于匹配策略融合的低误差快速图像拼接算法  

Low-error and fast image stitching algorithm based on merging match strategies

作　　者：杨明东 石英[1] 华逸伦[1] 朱剑怀 Yang Mingdong;Shi Ying;Hua Yilun;Zhu Jianhuai(School of Automation,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学自动化学院,武汉430070

出　　处：《计算机应用研究》2019年第4期1222-1227,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(61673306);江苏省重点研发计划资助项目(BE2016155)

摘　　要：针对目前多数拼接算法正确匹配率低、耗时及误差大的问题,提出了基于匹配策略融合的改进图像拼接算法。首先仅在图像的重合区域提取SIFT特征点,并计算SURF特征描述符;其次,融合改进的最近邻比次近邻、双向交叉检查及匹配差值的阈值化三种匹配策略,结合坐标约束及RANSAC算法完成特征匹配;最后,提出利用配准参数计算任意图像到参考图像坐标空间的投影变换模型的方法,并利用多线程技术将所有图像投影至参考坐标系,经过亮度校正、加权融合后合成全景图。实验结果表明,提出的拼接算法正确匹配率提高了10%～20%,拼接总耗时约为传统逐帧扩大式拼接算法的1/3,且累计误差大大降低,拼接图像畸变小。To deal with low correct matching,high time-consuming and high-error in most stitching algorithms,this paper developed an improved image stitching algorithm based on merging match strategies.First,the algorithm extracted SIFT feature points only from the overlap area of pair-image,and computed the SURF descriptors.Second,it merged three match strategies included ratio of the nearest neighbor and the 2nd nearest neighbor distances,cross-check of two directions and pair-matches'difference threshold method,combined coordinate constraint with RANSAC algorithm to complete the feature match.Finally,it used registration data to calculate the projection model which could transform any image in the sequences to reference image coordinate space,then used multithread technology to accelerate.After the brightness correction,the algorithm used the weighted fusion method to composite panorama images.The experimental results show that the proposed algorithm obtains a 10%~20%correct rate increase in feature match with about 1/3 time-consuming of the traditional frame-to-frame expand stitching algorithm,greatly decreases in accumulated error,and little image distortion.

关 键 词：匹配策略融合 快速图像拼接 累计误差 投影变换模型 多线程 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]