Comparative Study of Spectral Reconstruction Algorithms Based on Asymmetric Influence Factors

Spectral reconstruction is an important way to acquire a high-spatial-resolution multispectral image, and the spectral reconstruction algorithm is the key to its implementation. This work conducts a comparative study of spectral reconstruction algorithms under different asymmetric influencing factors to provide an overview of their performance. Seventeen spectral reconstruction algorithms with different mathematical bases were implemented and compared regarding their adaptability to response format, imaging noise, spectra type, exposure change, and the quality of the reconstructed images. Based on the principles and characteristics of the algorithm, qualitative and quantitative statistical analyses of the results were carried out. Results show that most of the current algorithms: (1) are adaptive to raw and image signal processing (ISP) responses for spectral reconstruction, (2) decrease their spectral reconstruction accuracy with an increase in imaging noise, (3) give poor performance in reconstructing smooth spectra using non-smooth spectra, and (4) exhibit different degrees of sensitivity to exposure changes. In addition, the image quality reconstructed using the raw response is superior to the ISP response.