Implementation of Wiener Algorithm for Spectral Reflectance Reconstruction of 3 and 6 Channel Images

Abstract

Wiener estimation is a widely used technique for the spectral reflectance reconstruction of colored objects. In our study the reflectance estimation of 3-channel images acquired using traditional RGB camera was compared with that of 6-channel images captured by a modified 6-channel camera. 240 patches of ColorChecker DC Chart were used for training and testing of the models.Their performance was assessed via quality metrics PSNR, RMSE and CIE DE2000 based on the reconstructed and spectrophotometrically measured reflectance values. Compared to the 3-channel models, 6-channel models in general provided better results. Lowest color difference CIE DE2000 was obtained when using 10 calculation terms in the 3-channel algorithm and 13 terms in the 6-channel. Best reconstruction models, i.e. lowest CIE DE2000 and RMSE and highest PSNR values, were found for the skin and neutral tones while the performance was the poorest with the saturated color patches. The obtained results can be of use in various practical applications, such as in a printing workflow with high precision color rendering or for accurate digitization of paintings and other works of art.

Keywords

• color accuracy
• machine learning
• spectral reflectance reconstruction
• Wiener estimation