Enhancing Spectral Color Images by RGB-Based Sharpening

摘要

This work examines a possibility to enhance the spatial resolution of spectral color images by using RGB color images. Nu-anceFX spectral imaging system, which allows the measuring of both spectral and RGB color images at different resolution levels using the identical geometry, is used for acquiring the images needed. The used enhancing process is based on the correspondence between spectra of a low spatial resolution spectral image, LRspe. and RGB-triplets of a high spatial resolution RGB image, HR_(RGB), both globally and locally. In a global approach, an estimated spectrum corresponding to index number i is defined as the average of all the spectra mapping into that index number, whereas in a local approach an estimated spectrum is defined as a spectrum of the closest mother pixel with the same index number. In the cases where the index number of pixel to be estimated equals to zero, the missing spectrum is estimated based on its neighborhood. The estimated high resolution spectral images, EHRspe. are evaluated by comparing them to the real, measured high resolution spectral images, HRspe, with respect to RMS errors and ΔE_(ab)~* color differences under four different CIE illumi-nants, D_(65), A, F8 and F11.