Spectral Reflectance Reconstruction from Camera Responses by Adaptively Constructing Weight Matrix Combining Colorimetric and Lightness Similarities

An improved method for spectral reflectance reconstruction from digital camera raw RGB responses of pixels is proposed by adaptively weighting training samples considering colorimetric and lightness similarities. The proposed method was based on an adaptive local weighted linear regression model by using a Gaussian function in weighting matrix construction. The novelty of our method is designing the weighting matrix combining colorimetric and lightness similarities. The proposed method was tested using two different standard color charts, with a simulated digital camera based on the camera spectral sensitivity. Experimental results indicate that the proposed method exhibits considerable improvements in terms of the spectral reflectance and the colorimetric values in comparison with existing methods.