Research was performed to design an LED-based spectral imaging system having channels, commonly referred to as a multispectral imaging system. The first part tackled the evaluation of a camera model in predicting the signals of a 10 LED LEDmotive Technologies Spectra Tunelab coupled with a Finger Lakes Instrumentation panchromatic camera. The camera model was shown to be valid and effective in predicting the camera signal taking into account the color transformation noise. The second part involved the computational selection of 10 LEDs in order to determine the optimum combination for a custom Spectra Tunelab. The computational selection used the spectral data provided by the manufacturer for their 37 available LEDs. The LEDs were grouped according to a specified wavelength range. The binning process helped in decreasing the computational cost and time; the possible combinations were reduced to 110,592 from the initial calculated value of 348,330,136 possible combinations. The combinations were further reduced to 1000 according to spectral reflectance Root-Mean-Square-Error (RMSE). The Euclidean and score ranking methods were then used to evaluate color transformation noise, spectral error and colorimetric accuracy. Goodness of Fit Coefficient and Throughput were calculated as well to further evaluate the combinations. A compromise among the values were reached to identify the best possible LED combination. The optimal combination has peak wavelengths at 390 nm, 450 nm, 475 nm, 505 nm, 540 nm, 550 nm, 590 nm, 620 nm, 660 nm, and 745 nm. All the LEDs were narrow band except the LED with its peak wavelength at 550 nm. This particular LED was similar to the human visual system’s luminous efficiency function. Its inclusion was important for colorimetric accuracy and small color transformation noise. When evaluating a large color-gamut target made using commonly used commercial pigments and several artist pigments, the following quality metrics were achieved: average ∆E00 of 0.12, total Noise, N of 3.35, a lightness noise (∆L) of 1.22, spectral reflectance RMSE of 6.4 x10-3, GFC of 0.97 and a total throughput of 646.85.
Library of Congress Subject Headings
Multispectral imaging; Light emitting diodes; Colorimetry
Color Science (MS)
Roy S. Berns
Paray, Jenibel N., "LED Selection for Spectral (Multispectral) Imaging" (2020). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus