[vslist] Summary: hyperspectral image question from 4 February 2004

[Alexa I. Ruppertsberg]

Hello,

First of all: Thank you very much for all your responses! I was
overwhelmed by so many helpful emails!

Yes, there is a lot of info out there on Google! (I had searched 'Web of
Science'!!). In the future I shall ignore 'Web of Science' and turn to 
Google for all my questions and the CVNet- and vslists! :-)

I have compiled a summary of your responses, which includes citations
from your emails (which I have not marked as such, I just tell you now!)
to explain the concept of hyperspectral images. It also contains
references and webpage links that you might find useful.

This summary does not claim to be exhaustive and I have not included all
links and suggestions I received. Please, don't take this personally.

Thanks again!

Alexa
P.S. Sorry if you receive this email twice.

-----------------------------------------------------------------
Summary of Responses to Hyperspectral Image Question, posed on 4.2.2004 
on vslist
-----------------------------------------------------------------

There are multi-, hyper- and ultra-spectral images. An explanation of 
the terms can be found here (http://www.fas.org/irp/imint/hyper.htm , 
http://www.techexpo.com/WWW/opto-knowledge/why_spectral.html ).

Multi: including other bands than visible band, like IR and UV etc
Hyper: usage of narrow bands
Ultra: usage of even narrower bands

The basic concept for a visual scientist:
The idea is to measure an entire image using narrow-band filters, i.e. 
ones, which pass only a small band of wavelengths. Usually the method is 
to fix a digital camera on a stand, and then take pictures of the scene 
through a set of filters, which vary in their waveband only (Filter 1: 
400-410 nm, Filter 2: 410-420 nm etc. The filters are usually selected 
for the visible band only). Thus, it takes some time to acquire a single 
set of pictures for a given scene (= hyperspectral image), which means 
that objects should be stationary.
To retrieve the spectrum of a single pixel in the hyperspectral image, 
one selects the corresponding pixel from all planes (i.e. different 
wavebands). This spectral signature (= spectrum) can then be multiplied 
by the appropriate receptor-sensitivity curve (of the camera, I believe) 
at the wavelength at which the image was acquired and can then be 
reconstructed.
Another application is to multiply the spectrum with cone-sensitivity 
curves to yield the corresponding L-, M-, and S-cone excitations, and 
similarly colour matching functions to yield X, Y and Z tristimulus 
values (from which RGB values can be computed).
The advantage of the hyperspectral solution is that it avoids any 
"transformation" of one tristimulus set to another (eg camera phosphors 
to TV phosphors). Transformations like that are only approximations and 
can lead to serious errors for colour pairs called "metamers". But with 
hyperspectral data, any set of tristimulus values can be selected and 
there is no metamerism problem.


Technical issues:
If the width of the filter is too narrow, or the lighting too dim, or 
the exposure for each image too short, the camera won't register 
anything. So the bandwidth has to be adjustable. Probably 10nm is as 
fine as is possible in daylight with long exposures.


References:

Chang, Chein-I.(2003). Hyperspectral imaging. Kluwer Academic/Plenum 
Publishers, New York. ISBN 0-306-47483-2 (£90!)

Chiao C-C, Cronin TW & Osorio D, Color signals in natural scenes: 
characteristics of reflectance spectra and effects of natural 
illuminants, JOSA A 17, 218-224 (2000).

Kraft, JM & Brainard, DH (1999). Mechanisms of color constancy under 
nearly natural viewing. Proceedings of the National Academy of Sciences, 
96, 307-313. (see Appendix)

Nascimento, S.M.C., Ferreira, F., and Foster, D.H. (2002). Statistics of 
spatial cone-excitation ratios in natural scenes. Journal of the Optical 
Society of America A, 19, 1484-1490

Ruderman DL, Cronin TW & Chiao C-C, Statistics of cone responses to 
natural images: implications for visual coding, JOSA A 15, 2036-2045 
(1998).


Webpages for downloading of hyperspectral images and more explanation:
·http://personalpages.umist.ac.uk/staff/david.foster/Hyperspectral_images_of_natural_scenes_02.html 

·http://psy197.psy.bris.ac.uk/hyper/
·http://color.psych.upenn.edu/hyperspectral/
·http://color.psych.upenn.edu/simchapter/simchapter.pdf
·http://www.cis.rit.edu/mcsl/research/CameraReports.shtml
(It explains the acquisition system and several spectral reconstruction 
methods. 8 different parts of a long report, looks interesting.)
·http://www.tsi.enst.fr/publications/Hardeberg/these.pdf
(A thesis (in English), looks good, but > 230 pages!)



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr. Alexa I. Ruppertsberg
Department of Optometry
University of Bradford
Bradford
BD7 1DP
UK
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~