Advanced dynamic weight-optimized semi-automatic 3D registration of SLT infrared datasets spanning 17 years: A case study on a complex-surfaced ancient bronze helmet

This study proposes a novel semi-automatic registration framework that performs three-dimensional registration of two infrared datasets acquired under solar loading thermography (SLT) on a complex-surfaced ancient bronze helmet. The datasets are collected on two days that are 17 years apart, from different viewing angles under natural light excitation. The data pre-processing sequence includes fast iterative filtering for denoising, principal component analysis for dimensionality reduction, and optical distortion correction. The proposed framework employs a dynamic weight-optimized perspective transformation that integrates manually selected matching points with residual-based real-time iterative weight optimization. This approach ensures robust registration while enhancing efficiency. The final results indicate that the method effectively compensates for viewpoint variations under complex conditions, ensuring accurate feature registration of the datasets. Such precise registration provides researchers with a reliable foundation for detailed analysis of thermal anomalies and material degradation.