Research on Ink Color Matching Based on the Integration of the NPE-Transformer Model and Kubelka-Munk Single Constant Theory

This paper proposes a novel approach that integrates the Kubelka-Munk (K-M) single constant theory with the NPE-Transformer model, incorporating adaptive nonlinear position encoding, for ink color matching prediction. The model takes spectral reflectance data and CMYK dot area percentages as inputs and outputs. Through the use of adaptive nonlinear position encoding, the model automatically learns wavelength positional features, enhancing its ability to capture spectral details. Additionally, the absorption and scattering coefficients from the K-M single constant theory are integrated into the loss function, reinforcing the model's understanding of physical principles and improving prediction accuracy. Experimental results show that the traditional Transformer model has an average prediction error of 9.31 %, while the NPE-Transformer model, combined with K-M theory, reduces the error to 5.27 %, demonstrating the potential and advantages of this approach for color matching prediction in the printing industry.