Reflectance spectra recovery from a single RGB image by adaptive compressive sensing

A new adaptive compressive sensing-based method is proposed for recovering spectral reflectance from a single RGB image by using dynamic principal components that are calculated from the dynamic eigenvector subspace. The dynamic eigenvector subspace is formed by its similarity to the RGB trichromatic values of the original and not just the neighbors to the target. Unlike recovering spectral reflectance using fixed principal components, the novelty of the proposed method is the use of the dynamic principal component by applying an adaptive compressive sensing method under the dynamic eigenvector subspace. The effectiveness of the proposed and the existing methods are compared by reviewing the spectral recovery accuracy under different illuminants and cameras. As the spectral recovery results show, the proposed method is an effective spectral reflectance recovery method. (letter)