Recovery of underwater object spectra from colored image
Abstract
We present a model for estimating the color signal and reflectance spectra of underwater samples from colored images using singular value decomposition. Simulated reconstructions of color signal and reflectance spectra agree well with measured color signals and calculated reflectances. Spectral reconstructions from image have the highest fidelity values for reconstructed color signals using 3 basis spectra ranging from 0.877 to 0.999. On the other hand, spectral reconstructions using 6 basis spectra have the lowest fidelity values ranging from –148.020 to –6.645. Discrepancies between simulated and image-based reconstructions are suggestive of deficiencies in camera correction or in the color signal model.