use of hyperspectral imaging for artwork evaluation
TRANSCRIPT
Use of hyperspectral imaging for artwork evaluation
Adam Polak1, Professor Stephen Marshall2, Dr David J. M. Stothard3
1 – Electronic and Electrical Engineering, University of Strathclyde, Glasgow ([email protected])
2 – Electronic and Electrical Engineering, University of Strathclyde, Glasgow
3 - Fraunhofer Centre for Applied Photonics, Fraunhofer UK Research Ltd
Hyperspectral Imaging
Initial classification results
Hyperspectral data cube is a stack of images in which each pixel
(vector) is represented by a spectral signature that characterizes
the underlying objects
Set of signal processing techniques is able to classify captured
objects based on their spectral signatures
Variety of hyperspectral imagers is available on the market,
characterised by different scanning methods and spectral ranges.
Firefly Imager - laser-based, mid-IR, active hyperspectral imaging
system is used in this project
Despite observed problems, classification was performed on available data
Intensity variation results in very poor classification performance
Artwork data collection Set of test paintings
with chosen paint
samples were
imaged in in the
near and mid
infrared spectral
range (from 1.5µm
to 3.7µm)
The outcomes of this project will guide the design and implementation of the next generation of hyperspectral
imagers for applications in the art world and beyond.
Nature Photonics 3, 627 - 629 (2009)
Intensity variation issue Beam deflection from the Imager scanning system caused changes in
collection efficiencies of reflected radiation. It has direct impact on spectral
signatures of imaged objects in different areas of the image
Low ---------- accuracy ---------- HighTraining set Classification results
Line-scan approach
Horizontal deflection of the beam caused the
biggest variation in collected signal therefore
new scan method was proposed, where
horizontal beam scan was replaced by movement
of the painting in front of the Imager performing
vertical line scan.
Noise reduction electronics Basic methodology for noise
reduction is to measure the
outgoing pulse energy (channel
B) and use that to normalise the
returned signal (channel A) from
the target.
Any jitter in the emitted energy
will be eliminated through the
A/B normalisation.
Pulse-by-pulse jitter
Normalised signal