随着计算机视觉技术的快速发展，获取广角度、高清晰图像的研究成为近年图像处理的热点问题。在日常生活中，单摄像头拍摄的图像存在拍摄死角、视野偏小的问题，而如果通过调焦和使用广角镜头来获得大场景图像则又会导致分辨率降低以及图像畸变，宽视野和高分辨率一直很难两全。因此，为了得到宽视野高分辨率的图像，可以通过对多摄像头从不同方向采集到的图像进行全景图像拼接。全景图像拼接技术已在航空航天、遥感图像、医学影像等领域得到了广泛应用。本文对于全景图像拼接中图像配准和图像融合两项关键技术中存在的问题进行了研究。

针对在图像发生模糊、视角变换、亮度变化及旋转缩放情况下传统ORB算法提取特征点数量不稳定，特征点匹配准确率低的问题，提出了一种改进的ORB算法。首先对图像进行网格化并提取特征点，接着引入四叉树结构使得提取到的特征点能够均匀分布，解决了图像中特征明显区域导致的提取特征点过于集中且增加后续匹配耗时的问题。然后，利用双向KNN算法对特征点进行粗匹配，并通过RANSAC算法剔除部分误匹配对。最后，通过高斯函数对网格加权的方式改进网格运动统计算法，进一步剔除误匹配对。实验结果表明，本文改进算法提高了算法特征点匹配精度。

针对传统全景图像融合中图像重叠部分存在运动物体而导致最终拼接结果中出现鬼影的问题，通过最佳缝合线算法来改进传统的拉普拉斯金字塔融合算法，消除了鬼影模糊的现象。实验结果表明，本文改进的融合算法取得了较好的效果。

针对传统的全景图像拼接结果中存在的倾斜弯曲问题，本文采用端到端对齐的方式，通过坐标系之间的转换来计算出图像坐标系与世界坐标系之间的偏差，最后将所有的累计误差平均分配到每张图像上，从而达到全景图像校正的目的。实验结果表明，此方法能够有效地削弱全景图像的倾斜弯曲问题。

With the rapid development of computer vision technology, the research on obtaining wide-angle and high-definition images has become a hot issue in image processing in recent years. In daily life, the images captured by a single camera have the problems of shooting dead angle and small field of vision. If the large scene images are obtained by focusing and using a wide-angle lens, the resolution will be reduced and the image distortion will be caused. It is difficult to complete the wide field of vision and high resolution. Therefore, in order to obtain wide field of vision and high resolution images, panoramic image stitching can be carried out by collecting images from different directions from multiple cameras. Panoramic image mosaic technology has been widely used in aerospace, remote sensing images, medical imaging and other fields. In this paper, the problems existing in the two key technologies of image registration and image fusion in panoramic image mosaic are studied.

An improved ORB algorithm is proposed to solve the problem that the number of feature points extracted by traditional ORB algorithm is unstable and the accuracy of feature point matching is low in the case of image blurring, perspective transformation, brightness change and rotation scaling. Firstly, the image is gridded and the feature points are extracted. Then, the quad-tree structure is introduced to make the extracted feature points evenly distributed, which solves the problem that the extracted feature points caused by the obvious feature areas in the image are too concentrated and the subsequent matching time is increased. Then, the bidirectional KNN algorithm is used to roughly match the feature points, and RANSAC algorithm is used to eliminate some mismatch pairs. Finally, the grid motion statistical algorithm is improved by weighting the grid with Gaussian function, and the false matching pairs are further eliminated. Experimental results show that the improved algorithm improves the matching accuracy of feature points.

In view of the ghosting phenomenon in the final stitching result caused by the existence of moving objects in the overlapping part of the traditional panoramic image fusion. This paper improves the traditional laplacian pyramid fusion algorithm through the optimal suture algorithm, so as to eliminate the ghosting blur. The experimental results show that the improved fusion algorithm in this paper has achieved good results.

In view of the tilt bending problem in the traditional panoramic image stitching results, this paper uses the end-to-end alignment method to calculate the deviation between the image coordinate system and the world coordinate system through the transformation between the coordinate systems. Finally, all the cumulative errors are evenly distributed to each image, so as to achieve the purpose of panoramic image correction. Experimental results show that this method can effectively weaken the tilt bending problem of panoramic images.

[1] Wang Z, Yang Z. Review on image-stitching techniques[J]. Multimedia Systems, 2020, 26(04): 413-430.

[2] 章毓晋. 中国图像工程: 2020[J]. 中国图象图形学报, 2021, 26(05): 978-990.

[3] 丁颖, 刘延伟, 刘金霞, 刘科栋, 王利明, 徐震. 虚拟现实全景图像显著性检测研究进展综述[J]. 电子学报, 2019, 47(07): 1575-1583.

[4] Yunyun Dong, Weili Jiao, Tengfei Long, Guojin He, Chengjuan Gong. An Extension of Phase Correlation-Based Image Registration to Estimate Similarity Transform Using Multiple Polar Fourier Transform[J]. Remote Sensing, 2018, 10(11): 1719-1745.

[5] Rasmy Laila, Sebari Imane, Ettarid Mohamed. Automatic Sub-Pixel Co-Registration of Remote Sensing Images Using Phase Correlation and Harris Detector[J]. Remote Sensing, 2021, 13(12): 2314-2333.

[6] 朱明, 姚强, 唐俊, 张艳. 超图约束和改进归一化互相关方法相结合的图像配准算法[J]. 国防科技大学学报, 2019, 41(03): 50-55.

[7] Cui Zhongjie, Qi Wenfa, Liu Yuxin. A Fast Image Template Matching Algorithm Based on Normalized Cross Correlation[J]. Journal of Physics: Conference Series, 2020, 1693(1): 2163-2172.

[8] 李培, 姜刚, 马千里, 薛万峰, 杨伟华. 结合张量与互信息的混合模型多模态图像配准方法[J]. 测绘学报, 2021, 50(07): 916-929.

[9] 曾安, 王烈基, 潘丹, 黄殷. 基于FCN和互信息的医学图像配准技术研究[J]. 计算机工程与应用, 2020, 56(18): 202-208.

[10] Bay H, Tuytelaars T, Gool L V. SURF: Speeded up robust features[C]// Proceedings of the 9th European conference on Computer Vision - Volume Part I. Springer-Verlag, 2006: 404-417.

[11] 赵谦, 童申鑫, 贺顺, 尹怡晨. 改进的SURF-RANSAC图像匹配算法[J]. 计算机工程与设计, 2021, 42(10): 2902-2909.

[12] Rublee E, Rabaud V, Konolige K, et al. ORB: an efficient alternative to SIFT or SURF[C]// IEEE International Conference on Computer Vision, 2011: 2564-2571.

[13] Calonder M, Lepetit V, Strecha C, et al. BRIEF: Binary Robust Independent Elementary Features[J]. Computer Vision - ECCV 2010, 11th European Conference on Computer Vision, Heraklion, Crete, Greece, September 5-11, 2010, Proceedings, Part IV, 2010.

[14] Suárez Iago, Sfeir Ghesn, Buenaposada José M., Baumela Luis. BEBLID: Boosted Efficient Binary Local Image Descriptor[J]. Pattern Recognition Letters, 2020, 133(4): 366-372.

[15] Zhang Hang, Zhao Mingchang. Panoramic Image Stitching Using Double Encoder - Decoders[J]. SN Computer Science, 2021, 2(2): 81-93.

[16] Nie Lang, Lin Chunyu, Liao Kang, Liu Shuaicheng, Zhao Yao. Unsupervised Deep Image Stitching: Reconstructing Stitched Features to Images[J]. Computer Vision and Pattern Recognition, 2021.

[17] 王丽丽, 刘辉, 郭肇禄. 图像融合在域内的算法研究进展[J]. 计算机技术与发展, 2021, 31(11): 8-14.

[18] 甄媚, 王书朋. 可见光与红外图像自适应加权平均融合方法[J]. 红外技术, 2019, 41(04): 341-346.

[19] 卢鹏, 卢奇, 邹国良, 王振华, 侯倩. 基于改进SIFT的时间序列图像拼接方法研究[J]. 计算机工程与应用, 2020, 56(01): 196-202.

[20] Bing Sun, Shan-Shan Hou, Sheng Zeng, Xin Bai, Shu-Wen Zhang, Jing Zhang. 3D characterization of porosity and minerals of low-permeability uranium-bearing sandstone based on multi-resolution image fusion[J]. Nuclear Science and Techniques, 2020, 31(10): 117-136.

[21] 方贤勇, 张明敏, 潘志庚, 罗斌, 王鹏. 基于图切割的图像拼接技术研究[J]. 中国图象图形学报, 2007(12): 2050-2056.

[22] 赵潇洒, 陈西江, 班亚, 张丹丹, 徐乐先. 融合改进SURF和Cell加速的幂函数加权图像拼接方法[J]. 激光与光电子学进展, 2020, 57(24): 198-208.

[23] 杨晓莉, 蔺素珍. 一种注意力机制的多波段图像特征级融合方法[J]. 西安电子科技大学学报, 2020, 47(01): 120-127.

[24] 么鸿原, 王海鹏, 林雪原, 潘新龙. 遥感图像拼接中改进的图像预处理算法研究[J]. 计算机与数字工程, 2020, 48(02): 428-432.

[25] 吴亮红, 孙亮, 周博文, 张红强, 卢明. 基于伪高动态范围的图像拼接预处理方法[J]. 中南大学学报(自然科学版), 2019, 50(07): 1619-1625.

[26] Mur-Artal R, JD Tardós. ORB-SLAM2: An Open-Source SLAM System for Monocular, Stereo, and RGB-D Cameras[J]. IEEE Transactions on Robotics, 2017, 33(5): 1255-1262.

[27] 程向红, 李俊杰. 基于运动平滑性与RANSAC优化的图像特征匹配算法[J]. 中国惯性技术学报, 2019, 27(06): 765-770.

[28] 柳长安, 艾壮, 赵丽娟. 基于网格运动统计的自适应图像特征匹配算法[J]. 华中科技大学学报(自然科学版), 2020, 48(01): 37-40+54.

[29] 王春洁, 卫朝霞. 基于泊松方程的VR全景图像镶嵌拼接缝消除[J]. 计算机仿真, 2020, 37(12): 135-139.

[30] 朱宁宁. 柱面全景图像拼接方法的仿真分析[J]. 测绘学报, 2017, 46(04): 487-497.

[31] 刘尚, 李哲. 面向野外火灾监测的图像拼接技术研究[J]. 计算机与数字工程, 2020, 48(03): 653-657.

[32] 翟潘, 王平. 自适应维纳滤波在钢水红外图像去噪中的应用[J]. 红外技术, 2021, 43(07): 665-669.

[33] 艾加秋, 王非凡, 杨兴明, 史骏, 刘凡. 基于背景匀质性双边滤波的SAR图像斑点噪声抑制算法[J]. 遥感学报, 2021, 25(05): 1071-1084.

[34] 梁利利, 高楠, 李建军. 基于小波变换和均值滤波的图像去噪方法[J]. 计算机与数字工程, 2019, 47(05): 1229-1232.

[35] 李伟. 图像几何校正算法和反混叠综述[J]. 现代计算机, 2020(03): 83-87.

[36] 万余庆, 郑鹏, 徐凯磊. 无人机航摄图像总体亮度统一调整技术研究[J]. 测绘通报, 2019(S1): 310-313+317.

[37] 朱庆辉, 尚媛园, 邵珠宏, 尹晔. 局部特征及视觉一致性的柱面全景拼接算法[J]. 中国图象图形学报, 2016, 21(11): 1523-1529.

[38] Fischler M A, Bolles R C. Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography - ScienceDirect[J]. Readings in Computer Vision, 1987: 726-740.

[39] Jia-Wang Bian, Wen-Yan Lin, Yun Liu, Le Zhang, Sai-Kit Yeung, Ming-Ming Cheng, Ian Reid. GMS: Grid-Based Motion Statistics for Fast, Ultra-robust Feature Correspondence[J]. International Journal of Computer Vision, 2020, 128: 1580-1593.

[40] 孙浩, 王朋. 一种基于区域划分的改进ORB算法[J]. 北京航空航天大学学报, 2020, 46(09): 1763-1769.

[41] 杨弘凡, 李航, 陈凯阳, 李嘉琪, 王晓菲. 基于改进ORB算法的图像特征点提取与匹配方法[J]. 图学学报, 2020, 41(04): 548-555.

[42] 么鸿原, 林雪原, 王海鹏, 郭强. 针对高分辨率遥感图像的融合拼接算法[J]. 计算机工程与设计, 2019, 40(11): 3230-3235.

[43] 斯捷, 肖雄, 李泾, 户胜鸿, 毛玉星. 基于格林坐标和改进最佳缝合线的图像拼接技术[J]. 计算机应用, 2021, 41(S1): 230-236.