Reconstruction of Spectral Reflectance Based on Fusion Convolution Neural Network

An improved method for spectral reflectance reconstruction from the RGB response of the digital camera is proposed by deep convolution neural network. The proposed method learns a fusion mapping theory that represents the basic characteristics of spectral reflectance, including spectral nonlinear mapping and spatial similarity. The deep residual network greatly deepens the network structure and improves the performance of the model, enabling the recovery network to make full use of the sample structure information to accurately recover the spectral reflectance. The performance of the method is compared with several existing methods in the cases of the same experimental data set. Experimental results show that the proposed method has the best accuracy in estimating spectral reflectance and colorimetric values in most cases in comparison with existing methods.