Reconstruction algorithms for RGB-based coded aperture snapshot spectral imaging

In coded aperture snapshot spectral imaging (CASSI), replacing the monochrome detector with a Bayer RGB sensor provides a compact means of enriching the measurements via pixel-wise wavelength selectivity. This modification, however, requires reconstruction methods that are consistent with the resulting color-dependent spectral modulation. In this work, we develop two RGB-aware reconstruction methods, LRSDN-RGB and SPECAT-RGB, by redesigning key components of two representative paradigms: an unfolding-based method (LRSDN) and an end-to-end transformer-based method (SPECAT). Simulation results show that LRSDN-RGB and SPECAT-RGB achieve PSNR improvements of 1.79 dB and 1.91 dB over their monochrome counterparts, respectively. Real measurements captured using a prototype CASSI system further demonstrate the effectiveness of the proposed methods.