Reconstruction of multispectral images from spectrally coded light fields of flat scenes

摘要

We present a novel method to reconstruct multispectral images of flat objects from spectrally coded light fields as taken by an unfocused light field camera with a spectrally coded microlens array. In this sense, the spectrally coded light field camera is used as a multispectral snapshot imager, acquiring a multispectral datacube in a single exposure. The multispectral image, corresponding to the light field’s central view, is reconstructed by shifting the spectrally coded subapertures onto the central view according to their respective disparity. We assume that the disparity of the scene is approximately constant and non-zero. Since the spectral mask is identical for all subapertures, the missing spectral data of the central view will be filled up from the shifted spectrally coded subapertures. We investigate the reconstruction quality for different spectral masks and camera parameter sets optimized for real life applications such as in-line production monitoring for which the constant disparity constraint naturally holds. For synthesized reference scenes, using 16 color channels, we achieve a reconstruction PSNRmathrm{PSNR} of up to 51?dB.