AIC 53rd Annual Meeting

Both multiband and multispectral imaging can provide a wealth of information about material characteristics and condition—with insights derived from qualitative and quantitative comparisons of images captured at different wavelengths and with different excitation sources. Workflows for these types of imaging often require costly additions to existing setups: IR-modified and/or monochrome cameras, filter sets, apochromatic lenses, and even licenses for proprietary image processing software, the sum of which can present a significant cost barrier. Certain equipment is essential, such as modified UV-VIS-IR full spectrum color or monochrome cameras. However, it is possible to perform multiband and multispectral imaging without the added cost of an apochromatic lens—one which produces a single focal plane across all wavelengths. The main challenge, however, with using a regular (achromatic) lens is the need to re-focus for each filter band, leading to registration issues across the entire set of images captured. This misalignment must be corrected post-capture not only to remove visual inconsistencies in false-color images but also to carry out any further computational analysis, such as Principal Component Analysis or Spectral Angle Mapping. 

With this issue in mind, this project has focused on developing a low-cost, open-source method for automating the registration of image sets generated from multiband and multispectral imaging workflows. Drawing on research beyond the field of art conservation, we have adapted Python code from a recent publication on vision-based robotics grasping in order to identify the specific feature-based pixel coordinates necessary for image registration. Specifically, the code utilizes an Open Source Computer Vision Library (OpenCV) tool called template matching as an alternative to feature-point detection algorithms or more complex object-detection models. In total, this method requires the addition of only a few printed paper targets and is designed to be integrated easily into existing multiband and multispectral imaging workflows. The current iteration of our adapted Python code can be executed directly from a computer’s command line, and we are hoping to create an ImageJ/FIJI plugin to make the script more readily available and user-friendly.