📄 中文摘要
针对历史板画的艺术科技调查，多模态图像数据（包括可见光摄影、红外反射成像、紫外荧光摄影、X射线照相和宏观摄影）的获取是关键步骤。为了进行全面分析，这些多模态图像需要实现像素级别的精确对齐，而目前这一过程仍多依赖于耗费大量人力的手动操作。多模态图像配准技术能够显著减轻这种繁重的手动工作，并为后续的自动化分析奠定基础。现有配准方法在处理涉及复杂非刚性变形的艺术品（如木板画）时，面临巨大挑战，尤其是在跨模态图像间缺乏足够的相似性特征时。利用板画上普遍存在的裂纹结构作为配准的内在特征，可以为解决这一难题提供有效途径。裂纹在不同模态图像中通常具有一定的可见度与独特的几何形态，可作为稳定的地标或特征点。通过构建一个粗到细的配准框架，首先在粗略阶段利用全局或稀疏的裂纹特征进行初步对齐，纠正较大尺度的几何失真。随后，在精细阶段，利用更密集的裂纹细节和局部几何信息，通过非刚性变换模型（如弹性变换或薄板样条）实现亚像素级的精确配准。这种方法能够有效应对木板因湿度、温度变化或老化导致的复杂局部变形，从而提升多模态图像间配准的鲁棒性和准确性。该技术对于历史文物的数字化保护、材料科学分析以及修复方案制定具有重要意义。
📄 English Summary
Coarse-to-Fine Non-rigid Multi-modal Image Registration for Historical Panel Paintings based on Crack Structures
Art technological investigations of historical panel paintings necessitate the acquisition of multi-modal image data, encompassing visual light photography, infrared reflectography, ultraviolet fluorescence photography, x-radiography, and macro photography. For comprehensive analysis, these multi-modal images demand pixel-wise alignment, a process still frequently performed manually and laboriously. Multi-modal image registration offers a substantial reduction in this manual effort and lays the groundwork for subsequent automated analyses. Existing registration methods face significant challenges when applied to artworks with complex non-rigid deformations, such as panel paintings, particularly due to the lack of sufficient similarity features across different modalities. Leveraging the ubiquitous crack structures found in panel paintings as intrinsic registration features provides an effective approach to overcome this difficulty. Cracks typically exhibit varying degrees of visibility and unique geometric forms across different image modalities, serving as stable landmarks or feature points. A coarse-to-fine registration framework can be constructed, wherein an initial alignment is performed in the coarse stage using global or sparsely distributed crack features to correct large-scale geometric distortions. Subsequently, in the fine stage, denser crack details and local geometric information are utilized to achieve sub-pixel accurate registration through non-rigid transformation models, such as elastic transformations or thin-plate splines. This methodology effectively addresses complex local deformations in wooden panels caused by humidity, temperature changes, or aging, thereby enhancing the robustness and accuracy of multi-modal image registration. This technology holds significant implications for the digital preservation of historical artifacts, material science analysis, and the development of conservation strategies.