During the fall we had the chance to work with several eminent art museums on the East Coast. The new RevealScan™100 hyperspectral imaging system was busy acquiring images from Cezanne, Renoir and other paintings. Tasks ranged from old to modern, detecting wear and degradation, authenticity of artworks, material problems, effects of repairs that may have been done hundreds of years ago, or are in process now. Humans enjoy art through vision mostly but there is a lot happening below the surface that we cannot see. These include substrate, structure, material, developing cracks, modifications, prior paintings, slight modifications by the artist, not to mention the most surprising detailed underdrawings. 

The tour was a lot of effort, both on Middleton’s part, but also on the part of the museum personnel to whom we are truly grateful. Thank you for the invaluable feedback on what the art curators and conservators community does, what they are interested in, what problems they are struggling with. Their valuable suggestions and requests are and will be incorporated in the current and future RevealScan™ models. 

Cultural heritage of course is much more than paintings, we have been scanning medieval chalices, jade artifacts, thousand-year-old statues, codex sheets, masks and many other very interesting, colorful, high-value objects. The system was easy to set up, usually took less than an hour and we were able to adapt the system to the varied tasks and problems. As we found out, each task is slightly different, at the same time new and exciting. We found out that flexibility and adaptability is highly valued by this community.

Hyperspectral imaging can provide powerful qualitative and quantitative information for a variety of samples. Plants, minerals, pharmaceuticals and even works of art can all benefit from information produced by a hyperspectral image. However, before we can begin successfully analyze these data, a series of preprocessing steps are needed in order to correct for variation in detector performance, structured noise and the illumination profile. A typical workflow for capturing a hyperspectral image includes acquisition of a sample file and at least two additional reference image files. The reference files include a white reference of a diffuse maximum reflective material such as porous Teflon ™ or Spectralon™ and a dark reference. The latter involves blocking any light from getting to the sensor in the camera. The dark reference can be two separate files such as a dark for the sample and dark for the white reference, with different integration times if necessary.

KemoQuant™ software from Middleton Spectral Vision has an automated step by step workflow for pre-processing these files to produce a robust hyperspectral image. Step one prompts the user to select the sample file that they want to preprocess. 

In step two, a dialog box pops up allowing the user to select the illumination type, the desired wavelength range for the preprocessed image and the ability to rotate or flip the image if required. Once this step is completed, the user is prompted to select the dark image for the sample, the white reference and dark image for the white reference in sequential order. Once all these files have been selected, the preprocessing step begins. The software begins to execute the equation below.

In addition, the images are analyzed for saturated pixels and if detected the entire pixel is removed. Depending on image file size and the computer being used, the preprocessing routine can be a few seconds to less than a minute. KemoQuant™ creates a unique file format for the preprocessed image with a *.s3d extension which additionally stores predefined color images so that the user can quickly start interrogating the spectral image. The integrity of the original raw image data is also maintained. Going forward, the user opens the preprocessed s3d file to work with the hyperspectral image for analysis.

References:

Preprocessing strategies to improve MCR analyses of hyperspectral images

Howland D.T. Jones ⁎ , David M. Haaland, Michael B. Sinclair, David K. Melgaard,

Aaron M. Collins, Jerilyn A. Timlin

Sandia National Laboratories, Albuquerque, NM 87185–0895, United States

Chemometrics and Intelligent Laboratory Systems

Volume 117, 1 August 2012, Pages 149-158