We analyze the performance of a multispectral system that works in the near- infrared (NIR) region of the electromagnetic spectrum (NIR: 800-1000 nm). The system, which uses a CCD camera as a sensor with five acquisition channels, is capable of reconstructing the NIR spectral reflectance curves for a wide range of samples with a high degree of accuracy. We carried out a study of the sources of error in the experimental system, developed a luminance adaptation model to remove the dependence of the captured images on the exposure time of the camera and the f-number of the objective lens, and performed reconstructions of the spectral reflectances of a set of 80 samples. We achieved the best results by using a 12-bit camera, considering a different luminance adaptation transform for each channel, and by using the pseudoinverse reconstruction method. Under these conditions, the system provided mean percentages of reconstruction higher than 99.8% and root-mean-square-error values lower than 0.17, and is therefore suitable for use as a spectrophotometric instrument.
© 2006 Optical Society of AmericaPDF Article