Spectral Reconstruction Based on Bayesian Regulation Neural Network

Abstract

In this paper, two sets of standard color cards of X-Rite are used to collect sample information using multi-spectral imaging system to study the spectral reconstruction accuracy. In order to solve the problems of the traditional neural network spectral reconstruction algorithms, this paper proposes appropriate improvements. The polynomial regression method is used to extend the camera response. The Bayesian regularization is used to improve the over-fitting phenomenon of the neural network. The combination of the two is used to improve the spectral reconstruction accuracy, and the spectral reconstruction results are discussed and evaluated. The experimental results show that the improved algorithm has significantly improved spectral accuracy and chromaticity accuracy, which is significantly higher than the traditional neural network spectral reconstruction algorithm. It has certain application value in the field of high spectral reconstruction accuracy and can meet the art. The requirement for high-precision color reproduction in the field.