Spectral prediction of CMYK printed images using stacking ensemble learning

Accurate spectral prediction of CMYK printed images is crucial for achieving precise and consistent color reproduction across various printing systems. Traditional physical models often fail to capture the nonlinear interactions between ink layers, while data-driven approaches, often relying on individual models, also face limitations in enhancing prediction accuracy. To address these challenges, we propose a novel Stacking Ensemble learning framework for Spectral Prediction (SESP) of CMYK printed images. The SESP model incorporates a two-layer stacking architecture, combining base regressors such as Support Vector Regression (SVR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost), with Ridge Regression serving as the meta-regressor. Comprehensive experiments demonstrated that SESP consistently outperforms both traditional physical models and existing data-driven approaches.