Multi- and Single-output Support Vector Regression for Spectral Reflectance Recovery

In this paper, we deal with the problem of reflectance recovery from multispectral camera output using Support Vector Regression (SVR). As standard, SVR is unidimensional, the spectral reflectance recovery requires a multi-dimensional output. We propose two ways of adaptation: the transformation of the dataset (camera output) to a scalar-valued composite data model on the one hand, and the adaptation of a recent multi-output SVR on the other hand. We compare both performances to a Wiener-based reflectance recovery. The results are quite satisfactory and the comparison points out the advantages and drawbacks of each one of the proposed methods.