Reflectance estimation using local regression methods

Regression methods have been widely used in the problem of spectral reflectance estimation from camera responses, due to their simple application without needing prior knowledge of the imaging system. These methods can be called global regression methods since the regression functions are trained on all the training samples. Recently, local learning methods have received considerable attention due to their capability in exploiting the local manifold structure of data. In this paper, we propose a set of reflectance estimation methods based on local regression methods. These methods can be seen as the local versions of the traditional global regression methods. The training set is confined to the test point's k-nearest neighbors. Experimental results show that the local ridge regression has the best generalization performance in the compared methods.