Micro LED可见光与光致发光图像亚像素快速配准

赵天元; 董登峰; 周维虎; 王国名

doi:10.37188/CO.2025-0142

Micro LED可见光与光致发光图像亚像素快速配准

doi: 10.37188/CO.2025-0142

cstr: 32171.14.CO.2025-0142

赵天元^{1, 2,},
董登峰^{1, 2},
周维虎^{1, 2, ,},
王国名^{1, 2}

1.
中国科学院微电子研究所, 北京 100029
2.
中国科学院大学, 北京 100049

基金项目: 国家重点研发计划资助项目（No. 2023YFF0719702）

详细信息

作者简介:
赵天元（1996—），男，河南焦作人，博士研究生，2018年于吉林大学获得学士学位，主要从事精密检测方面的研究。E-mail：zhaotianyuan@ime.ac.cn

周维虎（1962—），男，安徽无为人，博士，研究员，博士生导师，1983年、2000年于合肥工业大学分别获得学士和博士学位，主要从事光电检测、光电系统总体设计与集成测试、光电精密测量技术与仪器等方面的研究。E-mail： zhouweihu@ime.ac.cn

中图分类号: TP394.1;TH691.9
计量
- 文章访问数: 7
- HTML全文浏览量: 3
- PDF下载量: 0
- 被引次数: 0
出版历程
- 网络出版日期: 2026-02-11

Micro LED Visible Light and Photoluminescence Image Sub-pixel Fast Registration

ZHAO Tian-yuan^{1, 2
,},
DONG Deng-feng^{1, 2},
ZHOU Wei-hu^{1, 2
, ,},
WANG Guo-ming^{1, 2}

1.
Institute of Microelectronics of the Chinese Academy of Sciences, Beijing 100029, China
2.
University of Chinese Academy of Sciences, Beijing 100049, China

Funds: Supported by National Key R & D Program of China (No. 2023YFF0719702)

More Information

Corresponding author: zhouweihu@ime.ac.cn

摘要

摘要:
针对Micro LED缺陷检测中可见光（RGB）与光致发光（Photoluminescence, PL）图像因模态差异较大而难以实现高精度配准的难题，本研究致力于开发一种具备亚像素级精度和高鲁棒性的多模态图像配准方法，从而建立芯片物理结构与电学性能之间的映射关系。我们提出了一种结合结构特征约束与双向残差优化的配准方法：首先，基于Micro LED规则阵列的几何特性，设计了差异化的特征检测策略：在RGB图像中，通过椭圆拟合和基于密度的空间聚类算法（Density-Based Spatial Clustering of Applications with Noise, DBSCAN）精确提取电极中心；而在PL图像中，则采用改进的分水岭算法结合亚像素精修技术来定位芯片中心。其次，在配准优化阶段构建了双向残差约束框架，并引入基于残差分布的置信度加权机制，通过迭代重加权最小二乘法求解最优仿射变换参数。实验结果表明，本方法的平均绝对误差（Mean Absolute Error, MAE）为0.823像素，达到了亚像素级精度；与基线方法相比，MAE显著降低了94.2%。同时，均方根误差（Root Mean Square Error, RMSE）为0.996像素，最大误差（Max Error）控制在2.839像素以内，内点率达到75.0%，单次配准平均耗时仅为0.036秒，与互信息（Mutual Information, MI）等传统方法相比，运行效率实现了数量级提升。基于上述策略，本方法有效克服了多模态图像中的特征失配和异常点干扰问题，在配准精度、鲁棒性和效率方面均优于传统方法，为Micro LED芯片的精确缺陷检测与多模态分析提供了可靠的技术基础。
- Micro LED /
- 多模态图像配准 /
- 亚像素 /
- 双向残差约束 /
- 仿射优化
Abstract:
To address the challenge of achieving high-precision registration between visible light (RGB) and photoluminescence (PL) images in Micro LED defect inspection, which arises from substantial modality differences, this study introduces a robust multimodal image registration approach capable of attaining sub-pixel accuracy, aiming to establish a direct mapping between the physical structure and electrical characteristics of the chips. We propose a registration method that integrates structural feature constraints with bidirectional residual optimization. First, leveraging the geometric regularity of Micro LED arrays, a tailored feature detection strategy is employed: electrode centers in RGB images are accurately extracted via ellipse fitting and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), while chip centers in PL images are localized using an enhanced watershed algorithm with sub-pixel refinement. Second, during the registration optimization stage, a bidirectional residual constraint framework is constructed, incorporating a confidence weighting mechanism derived from residual distribution analysis. The optimal affine transformation parameters are then estimated using an iterative reweighted least squares method. Experimental results demonstrate that the proposed method achieves sub-pixel-level accuracy, with a mean absolute error (MAE) of 0.823 pixels, representing a 94.2% reduction compared to baseline methods. The root mean square error (RMSE) is 0.996 pixels, the maximum error remains below 2.839 pixels, and the inlier rate attains 75.0%. Each registration process takes only 0.036 seconds on average, achieving an order-of-magnitude improvement in computational efficiency over traditional mutual information (MI) methods. By effectively mitigating feature mismatch and outlier interference in multimodal images, the proposed method outperforms conventional approaches in terms of registration accuracy, robustness, and efficiency, thereby providing a reliable technical foundation for precise defect detection and multimodal analysis of Micro LED chips.
- Micro LED /
- multimodal image registration /
- sub-pixel /
- bidirectional residual constraint /
- affine optimization

HTML全文

图 1 RGB–PL图像配准流程

Figure 1. RGB–PL image registration workflow

下载: 全尺寸图片幻灯片

图 2 Micro LED实验平台

Figure 2. Micro LED experimental platform

下载: 全尺寸图片幻灯片

图 3 四组Micro LED图像配准结果

Figure 3. Registration results for four Micro LED image sets

下载: 全尺寸图片幻灯片

图 4 不同配准方法的实验结果

Figure 4. Experimental results of different registration methods

下载: 全尺寸图片幻灯片

图 5 高斯噪声鲁棒性测试结果

Figure 5. Robustness test results under Gaussian noise. (a) RMSE; (b) MAE; (c) Max Error; (d) Inlier Rate

下载: 全尺寸图片幻灯片

图 6 椒盐噪声鲁棒性测试结果

Figure 6. Robustness test results under salt-and-pepper noise. (a) RMSE; (b) MAE; (c) Max Error; (d) Inlier Rate

下载: 全尺寸图片幻灯片

图 7 旋转变换鲁棒性测试结果

Figure 7. Robustness test results under rotation transformations. (a) RMSE; (b) MAE; (c) Max Error; (d) Inlier Rate

下载: 全尺寸图片幻灯片

图 8 缩放变换鲁棒性测试结果

Figure 8. Robustness test results under scaling transformations. (a) RMSE; (b) MAE; (c) Max Error; (d) Inlier Rate

下载: 全尺寸图片幻灯片

表 1 本文方法的配准结果

Table 1. Registration results of the proposed method

	评价指标	数值	单位
几何精度	均方根误差(RMSE)	0.996	px
	平均绝对误差(MAE)	0.823	px
	最大误差(Max Error)	2.839	px
	内点率(Inlier Rate)	75.0	%
图像质量	结构相似性(SSIM)	0.334	-
图像质量	互信息(MI)	1.352	nats
运行效率	配准时间(Time)	0.036	s

下载: 导出CSV

表 2 四组Micro LED图像配准评价指标

Table 2. Registration evaluation metrics for four Micro LED image sets

组别（Group）	均方根误差 (RMSE)	平均绝对误差 (MAE)	最大误差 (Max Error)	内点率 (Inlier Rate)	结构相似性 (SSIM)	互信息 (MI)	时间 (Time)
（1）	0.832	0.742	1.907	86.8%	0.346	1.287	0.034
（2）	0.963	0.777	3.130	74.4%	0.348	1.286	0.035
（3）	1.071	0.884	3.684	57.6%	0.328	1.287	0.032
（4）	1.611	1.316	3.806	63.6%	0.362	1.343	0.043

下载: 导出CSV

表 3 不同配准方法的实验结果

Table 3. Experimental results of different registration methods

组别（Group）	方法	均方根误差 (RMSE)	平均绝对误差 (MAE)	内点率 (Inlier Rate)	结构相似性 (SSIM)	互信息 (MI)	时间 (Time)
（1）	ORB	1384.348	1288.305	-	0.002	0.009	8.150
	WLD	NaN	NaN	NaN	0.426	1.274	57.661
	MI	NaN	NaN	NaN	0.426	1.305	55.491
	Phase Correlation	NaN	NaN	NaN	0.515	0.065	3.853
	本文方法	0.832	0.742	86.8%	0.346	1.287	0.034
（2）	ORB	2.040	1.344	0.9%	0.705	0.295	8.151
	WLD	NaN	NaN	NaN	0.431	0.668	95.604
	MI	NaN	NaN	NaN	0.418	1.192	36.039
	Phase Correlation	NaN	NaN	NaN	0.470	0.367	3.915
	本文方法	0.963	0.777	74.4%	0.348	1.286	0.035
（3）	ORB	2.240	1.381	0.8%	0.376	0.150	7.995
	WLD	NaN	NaN	NaN	0.417	1.145	35.734
	MI	NaN	NaN	NaN	0.436	1.226	39.203
	Phase Correlation	NaN	NaN	NaN	0.511	0.040	3.844
	本文方法	1.071	0.884	57.6%	0.328	1.287	0.032
（4）	ORB	1207.783	1098.011	-	0.002	0.015	8.257
	WLD	NaN	NaN	NaN	0.440	1.107	96.789
	MI	NaN	NaN	NaN	0.464	1.239	45.807
	Phase Correlation	NaN	NaN	NaN	0.535	0.006	3.932
	本文方法	1.611	1.316	63.6%	0.362	1.343	0.043
注：表中“-”表示配准失败，无法计算内点率；“NaN”表示对应方法的配准机制不依赖特征点匹配，基于点对的几何精度指标（RMSE, MAE, 内点率）不适用。该类方法的性能通过图像质量指标（SSIM、MI）和配准时间进行评估。

下载: 导出CSV

表 4 消融实验结果

Table 4. Ablation study results

组别（Group）	均方根误差 (RMSE)	平均绝对误差 (MAE)	最大误差 (Max Error)	时间 (Time)
G0	39.538	14.311	172.762	0.004
G1	3.147	2.491	8.573	0.004
G2	1.690	1.394	4.614	0.008
G3	1.424	1.133	4.093	0.029
G4（Ours）	0.996	0.823	2.839	0.036

下载: 导出CSV

参考文献(30)

[1]	VELESACA H O, BASTIDAS G, ROUHANI M, et al. Multimodal image registration techniques: a comprehensive survey[J]. Multimedia Tools and Applications, 2024, 83(23): 63919-63947. doi: 10.1007/s11042-023-17991-2
[2]	林坚普, 王廷雨, 蔡苾芃, 等. 改进相位相关与聚类思想的Micro-LED芯片显微图像拼接[J]. 光学精密工程, 2025, 33(4): 641-652. doi: 10.37188/OPE.20253304.0641 LIN J P, WANG T Y, CAI B P, et al. Improved phase correlation and clustering ideas for Micro-LED chip microscopic image stitching[J]. Optics and Precision Engineering, 2025, 33(4): 641-652. (in Chinese). doi: 10.37188/OPE.20253304.0641
[3]	ZHAO G B, SU Y H, ZHU L H, et al. Micro-LED optical properties distribution research based on AOTF hyperspectral imager using empirical cumulative function[J]. Optics and Lasers in Engineering, 2024, 181: 108425. doi: 10.1016/j.optlaseng.2024.108425
[4]	谢军, 张威, 谢子敬, 等. 蜂巢表面AlGaInP基倒装Micro-LED的光电性能[J]. 发光学报, 2025, 46(10): 1953-1960. XIE J, ZHANG W, XIE Z J, et al. Photoelectric performance of AlGaInP-based flip-chip Micro-LEDs with honeycomb surface texture[J]. Chinese Journal of Luminescence, 2025, 46(10): 1953-1960. (in Chinese).
[5]	王伟, 张腾飞, 王绶玙. GaN基Micro-LED反向漏电流失效机理分析[J]. 发光学报, 2024, 45(9): 1539-1546. doi: 10.37188/CJL.20240164 WANG W, ZHANG T F, WANG SH Y. Analysis of failure mechanism of reverse leakage current in GaN-based Micro-LEDs[J]. Chinese Journal of Luminescence, 2024, 45(9): 1539-1546. (in Chinese). doi: 10.37188/CJL.20240164
[6]	ZHU J K, ZHENG B Y, XIONG B, et al. SynMSE: a multimodal similarity evaluator for complex distribution discrepancy in unsupervised deformable multimodal medical image registration[J]. Medical Image Analysis, 2025, 103: 103620. doi: 10.1016/j.media.2025.103620
[7]	ZHOU ZH Y, WANG SH K, HU J S, et al. Unsupervised registration for liver CT-MR images based on the multiscale integrated spatial-weight module and dual similarity guidance[J]. Computerized Medical Imaging and Graphics, 2023, 108: 102260. doi: 10.1016/j.compmedimag.2023.102260
[8]	WU J, CUI ZH M, SHENG V S, et al. A comparative study of SIFT and its variants[J]. Measurement Science Review, 2013, 13(3): 122-131. doi: 10.2478/msr-2013-0021
[9]	贾一鑫, 邓魏永, 殷强, 等. 基于改进SURF的增强现实图像匹配方法[J]. 计算机技术与发展, 2024, 34(1): 59-64. JIA Y X, DENG W Y, YIN Q, et al. Augmented reality image matching method based on improved SURF[J]. Computer Technology and Development, 2024, 34(1): 59-64. (in Chinese).
[10]	BANSAL M, KUMAR M, KUMAR M. 2D object recognition: a comparative analysis of SIFT, SURF and ORB feature descriptors[J]. Multimedia Tools and Applications, 2021, 80(12): 18839-18857. doi: 10.1007/s11042-021-10646-0
[11]	ZHU B, ZHOU L, PU S M, et al. Advances and challenges in multimodal remote sensing image registration[J]. IEEE Journal on Miniaturization for Air and Space Systems, 2023, 4(2): 165-174. doi: 10.1109/JMASS.2023.3244848
[12]	李茂月, 许圣博, 孟令强, 等. 基于逐点前进法的改进型点云配准方法[J]. 中国光学(中英文), 2024, 17(4): 875-885. LI M Y, XU SH B, MENG L Q, et al. An improved point cloud registration method based on the point-by-point forward method[J]. Chinese Optics, 2024, 17(4): 875-885. (in Chinese).
[13]	VELESACA H O, BASTIDAS G, ROUHANI M, et al. Multimodal image registration techniques: a comprehensive survey[J]. Multimedia Tools and Applications, 2024, 83(23): 63919-63947. (查阅网上资料, 本条文献与第1条文献重复, 请确认).
[14]	BHARATI S, MONDAL M R H, PODDER P, et al. Deep learning for medical image registration: a comprehensive review[J]. arXiv preprint arXiv: 2204.11341, 2022. (查阅网上资料, 请作者核对文献类型及格式是否正确).
[15]	PAN ZH T, WEN ZH W. Analysis of image registration algorithm RANSAC and its improved algorithm[J]. IOP Conference Series: Earth and Environmental Science, 2021, 783(1): 012138. doi: 10.1088/1755-1315/783/1/012138
[16]	余四豪, 盖绍彦, 达飞鹏. 面向点云配准的RANSAC假设评估度量方法研究[J]. 中国光学(中英文), 2025, 18(3): 487-498. YU S H, GAI SH Y, DA F P. Improved RANSAC hypothesis evaluation metrics for point cloud registration[J]. Chinese Optics, 2025, 18(3): 487-498. (in Chinese).
[17]	DENG W Y, DONG ZH, ZHANG J Y, et al. Binocular vision system integrated with transcranial magnetic stimulation automatic therapeutic apparatus[M]//DUAN B Y, UMEDA K, HWANG W. Proceedings of the Seventh Asia International Symposium on Mechatronics: Volume I. Singapore: Springer, 2020: 424-433.
[18]	刘聪, 董文飞, 蒋克明, 等. 基于改进分水岭分割算法的致密荧光微滴识别[J]. 中国光学, 2019, 12(4): 783-790. doi: 10.3788/CO.20191204.0783 LIU C, DONG W F, JIANG K M, et al. Recognition of dense fluorescent droplets using an improved watershed segmentation algorithm[J]. Chinese Optics, 2019, 12(4): 783-790. (in Chinese). doi: 10.3788/CO.20191204.0783
[19]	ZHANG D SH, HOU W H, GUO J, et al. Efficient subpixel image registration algorithm for high precision visual vibrometry[J]. Measurement, 2021, 173: 108538. doi: 10.1016/j.measurement.2020.108538
[20]	张正祺, 陈玉翀, 达飞鹏, 等. 基于双向条纹点云匹配的复杂纹理误差校正[J]. 中国光学(中英文), 2025, 18(5): 1086-1096. doi: 10.37188/CO.2025-0040 ZHANG ZH Q, CHEN Y CH, DA F P, et al. Error correction of complex texture objects based on bidirectional fringe projection point cloud matching[J]. Chinese Optics, 2025, 18(5): 1086-1096. (in Chinese). doi: 10.37188/CO.2025-0040
[21]	SINDAGI V A, YASARLA R, PATEL V M. Pushing the frontiers of unconstrained crowd counting: new dataset and benchmark method[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, IEEE, 2019: 1221-1231.
[22]	REZA A M. Realization of the contrast limited adaptive histogram equalization (CLAHE) for real-time image enhancement[J]. Journal of VLSI Signal Processing Systems for Signal, Image and Video Technology, 2004, 38(1): 35-44. doi: 10.1023/B:VLSI.0000028532.53893.82
[23]	DONG Y T, WANG ZH W, GUO W Y. Overview of edge detection algorithms based on mathematical morphology[C]. 2022 IEEE 6th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC), IEEE, 2022: 1321-1326.
[24]	HOU W F, LI D W, XU CH, et al. An advanced k nearest neighbor classification algorithm based on KD-tree[C]. 2018 IEEE International Conference of Safety Produce Informatization (IICSPI), IEEE, 2018: 902-905.
[25]	钟国舜, 刘秋佐, 李萌, 等. 动态扫描场景下GM-APD激光雷达点云高精度配准方法研究[J]. 中国光学(中英文), 2025, 18(5): 1076-1085. ZHONG G SH, LIU Q Z, LI M, et al. High-precision registration methods for GM-APD LiDAR point clouds in dynamic scanning scenarios[J]. Chinese Optics, 2025, 18(5): 1076-1085. (in Chinese).
[26]	DEISENROTH M P, TURNER R D, HUBER M F, et al. Robust filtering and smoothing with Gaussian processes[J]. IEEE Transactions on Automatic Control, 2012, 57(7): 1865-1871. doi: 10.1109/TAC.2011.2179426
[27]	LAW J. Robust statistics—the approach based on influence functions[J]. Journal of the Royal Statistical Society Series D: the Statistician, 1986, 35(5): 565-566. doi: 10.1016/s0169-7161(97)15005-2
[28]	BANERJEE A, DAS N, SANTOSH K C. Weber local descriptor for image analysis and recognition: a survey[J]. The Visual Computer, 2022, 38(1): 321-343. doi: 10.1007/s00371-020-02017-x
[29]	MAES F, COLLIGNON A, VANDERMEULEN D, et al. Multimodality image registration by maximization of mutual information[J]. IEEE Transactions on Medical Imaging, 1997, 16(2): 187-198. doi: 10.1109/42.563664
[30]	FOROOSH H, ZERUBIA J B, BERTHOD M. Extension of phase correlation to subpixel registration[J]. IEEE Transactions on Image Processing, 2002, 11(3): 188-200. doi: 10.1109/83.988953