Skip to main content
ARS Home » Pacific West Area » Albany, California » Western Regional Research Center » Healthy Processed Foods Research » Research » Publications at this Location » Publication #272176

Title: Methods for correcting morphological-based deficiencies in hyperspectral images of round objects

Author
item Haff, Ronald - Ron
item SARANWONG, SIRINNAPA - National Food Research Institute - Japan
item KAWANO, SUMIO - National Food Research Institute - Japan

Submitted to: Journal of Near Infrared Spectroscopy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/12/2011
Publication Date: 1/3/2012
Citation: Haff, R.P., Saranwong, S., Kawano, S. 2012. Methods for correcting morphological-based deficiencies in hyperspectral images of round objects. Journal of Near Infrared Spectroscopy. 19:431-441.

Interpretive Summary: NIR images of curved surfaces contain undesirable effects due to the shape of the sample. A computer program was developed to remove the variation in pixel intensity based directly on well known physical effects involving light reflection and intensity. The ideal result would be a uniform image (as is appropriate for a uniform sample). The three predominant principles investigated are the behavior of light reflected from Lambertian surfaces, the 1/R2 relationship between light intensity and distance from the source, and the variation in arc length along a circle as seen from the detectors. Neglecting effects at the outer edge, pixel intensity variation was reduced from 110/255 to 18/255, or from 43 % to 7 %. The same principle can be applied to samples with circular cross sections along a particular axis, which includes many agricultural commodities. Contributing factors to the remaining pixel intensity variation error in the corrected images include specular reflection, unintended ambient light and reflections from surfaces.

Technical Abstract: NIR images of curved surfaces contain undesirable artifacts that are a consequence of the morphology, or shape of the sample. A software correction was developed to remove the variation in pixel intensity in hyperspectral images of spherical samples generated on a linescan type imaging system. The correction is based directly on well known physical effects involving light reflection and intensity. The three predominant principles investigated are the behavior of light reflected from Lambertian surfaces, the 1/R2 relationship between light intensity and distance from the source, and the variation in arc length along a circle as seen from the detectors. The algorithm was tested using hyperspectral images of a uniform spherical teflon sample. Pixel intensity profiles and histograms were generated for the corrected images and evaluated to determine the effectivness of the algorithm based on the fact that the ideal result would be a uniform image (as is appropriate for a uniform sample). Results indicate that the algorithm effectively improves pixel intensity uniformity, although some variability remains. Contributing factors to the remaining pixel intensity variation in the corrected images include specular reflection, unintended ambient light and reflections from surfaces. The same principle can be applied to samples with circular cross sections along a particular axis, which includes many agricultural commodities.