Estimating reflectance from multispectral camera responses based on partial least-squares regression

摘要

In multispectral imaging systems, the accuracy of reflec-ntance estimation can be degraded by the nonlinearity in imagingnprocess, which is due to non-Gaussian distribution of the data andnnonlinear optoelectronic conversion function of the camera. To dealnwith nonlinearity, we propose to extend camera responses by high-norder polynomials and reduce the overfitting problem by partialnleast-squares (PLS) regression. Experiment shows that, in terms ofnboth spectral and colorimetric error metrics, the proposed methodnperforms better than Wiener estimation and ordinary polynomial re-ngression, and is similar to polynomial regression withnregularization.