基于单应性扩散约束的二步网格优化视差图像对齐

陈殷齐; 郑慧诚; 严志伟; 林峻宇

doi:10.16383/j.aas.c210966

基于单应性扩散约束的二步网格优化视差图像对齐

doi: 10.16383/j.aas.c210966

陈殷齐^{1, 2,},
郑慧诚^{1, 3, 4,},
严志伟^1,,
林峻宇^5,

1.
中山大学计算机学院广州 510006
2.
季华实验室新型显示技术与装备研究中心佛山 528000
3.
机器智能与先进计算教育部重点实验室广州 510006
4.
广东省信息安全技术重点实验室广州 510006
5.
复旦大学计算机科学技术学院上海 200438

基金项目: 国家自然科学基金 (61976231), 广东省基础与应用基础研究基金 (2019A1515011869), 广州市科技计划项目 (201803030029) 资助

详细信息

作者简介:
陈殷齐：中山大学计算机学院硕士研究生. 主要研究方向为图像对齐与拼接. E-mail: chenyq277@mail2.sysu.edu.cn

郑慧诚：中山大学计算机学院副教授. 2004年获得法国里尔第一大学博士学位. 主要研究方向为计算机视觉, 神经网络和机器学习. 本文通信作者. E-mail: zhenghch@mail.sysu.edu.cn

严志伟：中山大学计算机学院硕士研究生. 主要研究方向为深度学习, 目标检测. E-mail: yanzhw5@mail2.sysu.edu.cn

林峻宇：复旦大学计算机科学技术学院硕士研究生. 主要研究方向为深度学习, 具身智能. E-mail: 22210240210@m.fudan.edu.cn

计量
- 文章访问数: 435
- HTML全文浏览量: 271
- PDF下载量: 85
- 被引次数: 0
出版历程
- 收稿日期: 2021-10-13
- 录用日期: 2022-05-17
- 网络出版日期: 2022-09-05
- 刊出日期: 2024-06-27

Parallax Image Alignment With Two-stage Mesh Optimization Based on Homography Diffusion Constraints

CHEN Yin-Qi^{1, 2
,},
ZHENG Hui-Cheng^{1, 3, 4
,},
YAN Zhi-Wei^1
,,
LIN Jun-Yu^5
,

1.
School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou 510006
2.
New Display Technology and Equipment Research Center, Jihua Laboratory, Foshan; 528000
3.
Key Laboratory of Machine Intelligence and Advanced Computing, Ministry of Education, Guangzhou 510006
4.
Guangdong Province Key Laboratory of Information Security Technology, Guangzhou 510006
5.
School of Computer Science, Fudan University, Shanghai 200438

Funds: Supported by National Natural Science Foundation of China (61976231), Guangdong Basic and Applied Basic Research Foundation (2019A1515011869), and Science and Technology Program of Guangzhou (201803030029)

More Information

Author Bio:
CHEN Yin-Qi　Master student at the School of Computer Science and Engineering, Sun Yat-sen University. His research interest covers image alignment and stitching

ZHENG Hui-Cheng　Associate professor at the School of Computer Science and Engineering, Sun Yat-sen University. He received his Ph.D. degree from University of Lille 1, France, in 2004. His research interest covers computer vision, neural networks, and machine learning. Corresponding author of this paper

YAN Zhi-Wei　Master student at the School of Computer Science and Engineering, Sun Yat-sen University. His research interest covers deep learning and object detection

LIN Jun-Yu　Master student at the School of Computer Science, Fudan University. His research interest covers deep learning and embodied artificial intelligence

摘要

摘要: 目前, 在带有视差场景的图像对齐中, 主要难点在某些无法找到足够匹配特征的区域, 这些区域称为匹配特征缺失区域. 现有算法往往忽略匹配特征缺失区域的对齐建模, 而只将有足够匹配特征区域中的部分单应变换系数(如相似性变换系数)传递给匹配特征缺失区域, 或者采用将匹配特征缺失区域转化为有足够匹配特征区域的间接方式, 因此对齐效果仍不理想. 在客观事实上, 位于相同平面的区域应该拥有相同的完整单应变换而非部分变换参数. 由此出发, 利用单应变换系数扩散的思想设计了一个二步网格优化的图像对齐算法, 简称单应扩散变换(Homography diffusion warping, HDW)算法. 该方法在第一步网格优化时获得有足够匹配特征区域的单应变换, 再基于提出的单应性扩散约束将这些单应变换系数扩散到邻域网格, 进行第二步网格优化, 在保证优化任务简洁高效的前提下实现单应变换系数的传播与图像对齐. 相较于现有的针对视差场景图像对齐算法, 所提方法在各项指标上都获得了更好的效果.
- 图像对齐 /
- 视差场景 /
- 网格优化 /
- 匹配特征缺失区域
Abstract: At present, the main difficulty in the image alignment with parallax scene is in the areas that cannot find sufficient matching features. We call these areas featureless regions. Cutting-edge research on parallax image alignment neglects modeling of regions without matching features. Indirect methods such as transferring partial homography of regions with matching features to featureless regions or transforming featureless regions to regions with matching features have been popularly used, which, however, do not guarantee satisfactory results. In fact, image regions belonging to the same plane should possess the same homography. In this paper, a two-stage mesh optimization algorithm, homography diffusion warping (HDW), is designed by homography diffusion. In the first stage, homography coefficients of mesh cells in the image regions with matching features are obtained. Then we propagate these homography coefficients to adjacent cells to form homography diffusion constraints, and perform the second stage optimization of the mesh by enforcing the constraints on the premise of ensuring the simplicity and efficiency of the optimization task. Compared with existing image alignment algorithms, the method proposed in this paper achieves better results on all metrics.
- Image alignment /
- parallax scene /
- mesh optimization /
- featureless regions

HTML全文

图 1 本文算法与现有方法的图像对齐效果对比

Fig. 1 Comparison of image alignment effects between existing methods and the method proposed in this paper

下载: 全尺寸图片幻灯片

图 2 HDW对齐示意图

Fig. 2 The alignment process of HDW

下载: 全尺寸图片幻灯片

图 3 平面分割特性的分析

Fig. 3 Analysis of plane segmentation characteristics

下载: 全尺寸图片幻灯片

图 4 各图像对量化指标的直观对比

Fig. 4 Intuitive comparison of the quantitative indicators on all the image pairs

下载: 全尺寸图片幻灯片

图 5 Plant实例上的对齐结果对比

Fig. 5 Comparison of the alignment results on the Plant case

下载: 全尺寸图片幻灯片

图 6 Carpark实例上的对齐结果对比

Fig. 6 Comparison of the alignment results on the Carpark case

下载: 全尺寸图片幻灯片

图 7 Stationery实例上的对齐结果对比

Fig. 7 Comparison of the alignment results on the Stationery case

下载: 全尺寸图片幻灯片

图 8 Temple与Railtrack实例上的对齐结果对比

Fig. 8 Comparison of the alignment results on the Temple and Railtrack cases

下载: 全尺寸图片幻灯片

图 9 HDW仍然有提升空间的实例

Fig. 9 Examples where HDW still has room for improvement

下载: 全尺寸图片幻灯片

图 10 HDW的更多对齐结果实例

Fig. 10 Alignment results of HDW on more examples

下载: 全尺寸图片幻灯片

表 1 HDW相对其他算法在Err、PSNR和SSIM上的平均改进(%)

Table 1 The average improvement of HDW compared with other algorithms on Err, PSNR, and SSIM (%)

	SMH^[17]	PCPS^[18]	APAP^[6]	CPW^[8]	LLPC^[37]	ACW^[41]
Err	−63.80	−66.07	−67.15	−57.50	−71.78	−54.49
PSNR	+4.59	+9.01	+8.02	+7.48	+13.46	+7.12
SSIM	+6.24	+8.48	+8.65	+10.33	+36.45	+5.67

下载: 导出CSV

表 2 图像对的对齐效果量化指标对比

Table 2 Quantitative comparison of alignment performance on image pairs

		Plant	Carpark	Stationery
SMH^[17]	Err	1.2338	1.2659	1.0580
	PSNR	15.7496	12.3758	22.9365
	SSIM	0.6516	0.5719	0.9316
PCPS^[18]	Err	0.9396	1.1923	0.6695
	PSNR	15.3314	11.5550	23.5924
	SSIM	0.6858	0.5687	0.9355
APAP^[6]	Err	4.1854	1.1337	1.0154
	PSNR	13.2995	11.9361	23.4257
	SSIM	0.6245	0.6354	0.9236
CPW^[8]	Err	6.3718	1.6435	0.9038
	PSNR	12.7397	11.2034	23.4680
	SSIM	0.4818	0.5862	0.9258
ACW^[41]	Err	3.4302	0.7918	0.8070
	PSNR	13.2159	12.0738	23.6319
	SSIM	0.6589	0.6505	0.9278
HDW	Err	0.2741	0.2787	0.5134
	PSNR	18.1968	13.5019	24.2972
	SSIM	0.8221	0.7143	0.9400

下载: 导出CSV

表 3 其他算法相对HDW在耗时上的对比

Table 3 Temporal cost of HDW compared with those of other algorithms

方法	耗时占比 (%)	实际耗时 (ms)
HDW	100	106
SMH^[17]	1256	1331
PCPS^[18]	542	575
APAP^[6]	19351	20511
CPW^[8]	45	48
LLPC^[37]	298	316
ACW^[41]	20887	22140

下载: 导出CSV

参考文献(58)

[1]	Hartley R, Zisserman A. Multiple View Geometry in Computer Vision. Britain: Cambridge University Press, 2003.
[2]	Matsushita Y, Ofek E, Ge W, Tang X, Shum H Y. Full-frame video stabilization with motion inpainting. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006, 28(7): 1150−1163 doi: 10.1109/TPAMI.2006.141
[3]	韩敏, 闫阔, 秦国帅. 基于改进KAZE的无人机航拍图像拼接算法. 自动化学报, 2019, 45(2): 305−314 doi: 10.16383/j.aas.2018.c170521 Han Min, Yan Kuo, Qin Guo-Shuai. A mosaic algorithm for UAV aerial image with improved KAZE. Acta Automatica Sinica, 2019, 45(2): 305−314 doi: 10.16383/j.aas.2018.c170521
[4]	Li X, Hui N, Shen H, Fu Y, Zhang L. A robust mosaicking procedure for high spatial resolution remote sensing images. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 109: 108−125 doi: 10.1016/j.isprsjprs.2015.09.009
[5]	Xiang T Z, Xia G S, Bai X, Zhang L. Image stitching by line-guided local warping with global similarity constraint. Pattern Recognition, 2018, 83: 481−497 doi: 10.1016/j.patcog.2018.06.013
[6]	Zaragoza J, Chin T J, Tran Q H, Brown M S, Suter D. As-projective-as-possible image stitching with moving DLT. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(7): 1285−1298 doi: 10.1109/TPAMI.2013.247
[7]	Gao J, Kim S J, Brown M S. Constructing image panoramas using dual-homography warping. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Colorado Springs, USA: IEEE, 2011. 49−56
[8]	Zhang F, Liu F. Parallax-tolerant image stitching. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Columbus, USA: IEEE, 2014. 3262−3269
[9]	Lin K, Jiang N, Liu S, Cheong L F, Do M, Lu J. Direct photometric alignment by mesh deformation. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017. 2405−2413
[10]	Chang C H, Sato Y, Chuang Y Y. Shape-preserving half-projective warps for image stitching. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Columbus, USA: IEEE, 2014. 3254−3261
[11]	Lin C C, Pankanti S U, Ramamurthy K N, Aravkin A Y. Adaptive as-natural-as-possible image stitching. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE, 2015. 1155−1163
[12]	Xue W, Xie W, Zhang Y, Chen S. Stable linear structures and seam measurements for parallax image stitching. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(1): 253−261 doi: 10.1109/TCSVT.2021.3058655
[13]	Capel D. Image Mosaicing and Super-resolution. Berlin: Springer, 2004. 47−79
[14]	Szeliski R. Image Alignment and Stitching: A Tutorial, Technical Report MSR-TR-2004-92, Microsoft Research, Redmond, WA, USA, 2006.
[15]	Mann S, Picard R W. Video orbits of the projective group: A simple approach to featureless estimation of parameters. IEEE Transactions on Image Processing, 1997, 6(9): 1281−1295 doi: 10.1109/83.623191
[16]	Chang C, Chen C, Chuang Y. Spatially-varying image warps for scene alignment. In: Proceedings of the International Conference on Pattern Recognition. Stockholm, Sweden: IEEE, 2014. 64−69
[17]	Kim S, Uh Y J, Byun H. Generating panorama image by synthesizing multiple homography. In: Proceedings of the International Conference on Image Processing. Florida, USA: IEEE, 2012. 2981−2984
[18]	Zheng J, Wang Y, Wang H, Li B, Hu H M. A novel projective-consistent plane based image stitching method. IEEE Transactions on Multimedia, 2019, 21(10): 2561−2575 doi: 10.1109/TMM.2019.2905692
[19]	Lin W Y, Liu S, Matsushita Y, Ng T T, Cheong L F. Smoothly varying affine stitching. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Colorado Springs, USA: IEEE, 2011. 345−352
[20]	Igarashi T, Moscovich T, Hughes J F. As-rigid-as-possible shape manipulation. ACM Transactions on Graphics, 2005, 24(3): 1134−1141 doi: 10.1145/1073204.1073323
[21]	Guo Y, Liu F, Shi J, Zhou Z H, Gleicher M. Image retargeting using mesh parametrization. IEEE Transactions on Multimedia, 2009, 11(5): 856−867 doi: 10.1109/TMM.2009.2021781
[22]	Hu W, Luo Z, Fan X. Image retargeting via adaptive scaling with geometry preservation. IEEE Journal of Emerging and Selected Topics in Power Electronics, 2014, 4(1): 70−81
[23]	Chang C H, Chuang Y Y. A line-structure-preserving approach to image resizing. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Providence, USA: IEEE, 2012. 1075−1082
[24]	Wang Y S, Tai C L, Sorkine O, Lee T Y. Optimized scale-and-stretch for image resizing. ACM Transactions on Graphics, 2008, 27(5): Article No. 118
[25]	Li S, Yuan L, Sun J, Quan L. Dual-feature warping-based motion model estimation. In: Proceedings of the International Conference on Computer Vision. Santiago, Chile: IEEE, 2015. 428−4291
[26]	何川, 周军. 具有直线结构保护的网格化图像拼接. 中国图象图形学报, 2018, 23(7): 973−983 He Chuan, Zhou Jun. Mesh-based image stitching algorithm with linear structure protection. Journal of Image and Graphics, 2018, 23(7): 973−983
[27]	Li J, Wang Z, Lai S. Parallax-tolerant image stitching based on robust elastic warping. IEEE Transactions on Multimedia, 2018, 20(7): 1672−1687 doi: 10.1109/TMM.2017.2777461
[28]	Chen K, Tu J, Xiang B, Li L, Yao J. Multiple combined constraints for image stitching. In: Proceedings of the International Conference on Image Processing. Athens, The Hellenic Republic: IEEE, 2018. 1253−1257
[29]	李加亮, 蒋品群. 结合变形函数和幂函数权重的图像拼接. 计算机应用, 2019, 39(10): 3060−4064 Li Jia-Liang, Jiang Pin-Qun. Image stitching by combining deformation function and power function weight. Journal of Computer Applications, 2019, 39(10): 3060−4064
[30]	颜雪军, 赵春霞, 袁夏. 2DPCA-SIFT: 一种有效的局部特征描述方法. 自动化学报, 2014, 40(4): 675−682 Yan Xue-Jun, Zhao Chun-Xia, Yuan Xia. 2DPCA-SIFT: An efficient local feature descriptor. Acta Automatica Sinica, 2014, 40(4): 675−682
[31]	闫自庚, 蒋建国, 郭丹. 基于SURF特征和Delaunay三角网格的图像匹配. 自动化学报, 2014, 40(6): 1216−1222 Yan Zi-Geng, Jiang Jian-Guo, Guo Dan. Image matching based on SURF feature and Delaunay triangular meshes. Acta Automatica Sinica, 2014, 40(6): 1216−1222
[32]	Liu W X, Chin T J. Correspondence insertion for as-projective-as-possible image stitching. arXiv preprint arXiv: 1608.07997, 2016.
[33]	Fan B, Wu F, Hu Z. Line matching leveraged by point correspondences. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. San Francisco, USA: IEEE, 2010. 390−397
[34]	Joo K, Kim N, Oh T H, Kweon I S. Line meets as-projective-as-possible image stitching with moving DLT. In: Proceedings of the International Conference on Image Processing. Quebec, Canada: IEEE, 2015. 1175−1179
[35]	Xiang T, Xia G, Zhang L, Huang N. Locally warping-based image stitching by imposing line constraints. In: Proceedings of the International Conference on Pattern Recognition. Cancun, Mexico: IEEE, 2016. 4178−4183
[36]	Luo X, Li Y, Yan J, Guan X. Image stitching with positional relationship constraints of feature points and lines. Pattern Recognition Letters, 2020, 135: 431−440 doi: 10.1016/j.patrec.2020.05.003
[37]	Jia Q, Li Z, Fan X, Zhao H, Teng S, Ye X, Latecki L J. Leveraging line-point consistence to preserve structures for wide parallax image stitching. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. New York, USA: IEEE, 2021. 12186−12195
[38]	Herrmann C, Wang C, Bowen R S, Keyder E, Krainin M, Liu C, et al. Robust image stitching with multiple registrations. In: Proceedings of the European Conference on Computer Vision. Munich, Germany: 2018. 53−69
[39]	Lou Z, Gevers T. Image alignment by piecewise planar region matching. IEEE Transactions on Multimedia, 2014, 16(7): 2052−2061 doi: 10.1109/TMM.2014.2346476
[40]	Lee K Y, Sim J Y. Warping residual based image stitching for large parallax. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE, 2020. 8198−8206
[41]	Chen Y, Zheng H, Ma Y, Yan Z. Image stitching based on angle-consistent warping. Pattern Recognition, 2021, 117: Article No. 107993 doi: 10.1016/j.patcog.2021.107993
[42]	Song D Y, Um G M, Lee H K, Cho D. End-to-end image stitching network via multi-homography estimation. IEEE Signal Processing Letters, 2021, 99: 1−5
[43]	Le H, Liu F, Zhang S, Agarwala A. Deep homography estimation for dynamic scenes. In: Proceedings of the Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE, 2020. 7652−7661
[44]	Zhao Q, Ma Y, Zhu C, Yao C, Dai F. Image stitching via deep homography estimation. Neurocomputing, 2021, 450: 219−229 doi: 10.1016/j.neucom.2021.03.099
[45]	Dai Q, Fang F, Li J, Zhang G, Zhou A. Edge-guided composition network for image stitching. Pattern Recognition, 2021, 118: Article No. 108019 doi: 10.1016/j.patcog.2021.108019
[46]	Cao S, Hu J, Shen Z, Shen H. Iterative deep homography estimation. In: Proceedings of the 2022 Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE, 2022.
[47]	Zhao Y, Huang X, Zhang Z. Deep Lucas-Kanade homography for multimodal image alignment. In: Proceedings of the 2021 Conference on Computer Vision and Pattern Recognition. New York, USA: IEEE, 2021. 15950−15959
[48]	Mo Y, Kang X, Duan P, Li S. A robust UAV hyperspectral image stitching method based on deep feature matching. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1−14
[49]	Nie L, Lin C, Liao K, Liu S, Zhao Y. Unsupervised deep image stitching: Reconstructing stitched features to images. IEEE Transactions on Image Processing, 2021, 30: 6184−6197 doi: 10.1109/TIP.2021.3092828
[50]	Nie L, Lin C, Liao K, Liu S, Zhao Y. Depth-aware multi-grid deep homography estimation with contextual correlation. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(7): 4460−4472 doi: 10.1109/TCSVT.2021.3125736
[51]	Liu F, Gleicher M, Jin H, Agarwala A. Content-preserving warps for 3D video stabilization. ACM Transactions on Graphics, 2009, 28(3): 1−9
[52]	Fischler M A, Bolles R C. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Communication of ACM, 1981, 24(6): 381−395 doi: 10.1145/358669.358692
[53]	Zhang Z. Parameter estimation techniques: A tutorial with application to conic fitting. Image and Vision Computing, 1997, 15(1): 59−76 doi: 10.1016/S0262-8856(96)01112-2
[54]	Heckbert P S. Fundamentals of Texture Mapping and Image Warping [Ph.D. dissertation], EECS Department, University of California, USA, 1989.
[55]	Wang Z, Bovik A C, Sheikh H R, Simoncelli E P. Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing, 2004, 13(4): 600−612 doi: 10.1109/TIP.2003.819861
[56]	Lowe D G. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision, 2004, 60(2): 91−110 doi: 10.1023/B:VISI.0000029664.99615.94
[57]	Brown M, Lowe D G. Automatic panoramic image stitching using invariant features. International Journal of Computer Vision, 2007, 74(1): 59−73 doi: 10.1007/s11263-006-0002-3
[58]	Ma J, Jiang X, Fan A, Jiang J, Yan J. Image matching from handcrafted to deep features: A survey. International Journal of Computer Vision, 2021, 129: 23−79 doi: 10.1007/s11263-020-01359-2