Study on Spectral Reconstruction Based on Sample Optimization Method of Color Digital Camera

Abstract

According to the nonlinear characteristics of color digital camera imaging system, spectral reflectance reconstruction by the response values of color digital camera had been studied through the method of combining principal component analysis method with polynomial model under given conditions of illumination and observation environment. Sample optimization method selected standard color card sample as training sample which was similar to reconstruction of chroma space and broad representativeness in spectral space. It could avoid usual “over-fitting” problem brought by excessive samples in regression and could reconstruct spectral reflectance of object surface accurately. The results showed that polynomial model can simulate the nonlinear relationship accurately between the camera response and the coefficient vector obtained from the principal component analysis and the sample optimization could make full use of sample information so that the accuracy and stability of the regression function improved. The mean of RMSE of ColorChecker SG was 0.0247 and the average CIE DE2000 was 2.5123, the spectral and chroma accuracy have improved greatly compared with traditional algorithm.