September 26, 2021
Interesting paper from R. Pillay, J Hardeberg, and S George
Norwegian Institute of Science and Technology Read the Full Paper
Hyperspectral imaging has become an increasingly used tool in the analysis of works of art. However, the quality of the acquired data and the processing of that data to produce accurate and reproducible spectral image cubes can be a challenge to many cultural heritage users. The calibration of data that is both spectrally and spatially accurate is an essential step in order to obtain useful and relevant results from hyperspectral imaging. Data that is too noisy or inaccurate will produce sub-optimal results when used for pigment mapping, the detection of hidden features, change detection or for quantitative spectral documentation. To help address this, therefore, we will examine the specific acquisition and calibration workflows necessary for works of art. These workflows includes the key parameters that must be addressed during acquisition and the essential steps and issues at each of the stages required during post-processing in order to fully calibrate hyperspectral data. In addition we will look in detail at the key issues that affect data quality and propose practical solutions that can make significant differences to overall hyperspectral image quality.
In order to be able to obtain quantitative spectral data, a known reference must be used to normalize the acquired spectra to absolute reflectance factor. To do this, an acquisition must be made of a known reflectance standard or set of reflectance standards under the same operating conditions and acquisition parameters used for the acquisition of the work of art itself. These conditions include lighting levels, integration time, scan speed, acquisition geometry, camera gain settings etc.
Ideal reference targets are uniform and diffuse lambertian planar surfaces that have constant reflectance across a wide spectral range. These are typically durable and chemically inert PTFE (Polytetrafluoroethylene) targets, the most widely used of which is Spectralon® manufactured by Labsphere Inc.. Typically, a single reflectance standard is used and, indeed, the majority of the participants of the RRT acquired a single 99% reference standard image.
However, the quality of the radiometric calibration can often be improved by acquiring a set of different reference targets with a range of reflectivities. An example of such a target is shown. Again, these must be acquired using the same parameters as those for the painting acquisition either during the same scan sequence or separately.