Color imaging system high dynamic range colorimetric characterization modeling based on fusion kernel XGBoost.

Abstract

High dynamic range imaging exhibits lower resolution in both highly bright and deeply dark colors, which results in reduced accuracy in the measurement of photometry and colorimetry using color imaging systems. Based on the nonlinear capability of the Gaussian kernel function and the global linear trend of the linear kernel function, a Gaussian-linear fusion kernel is designed. Through multi-dimensional space mapped by the designed fusion kernel, a kernel XGBoost colorimetric characterization model is proposed. Combining fusion kernel and XGBoost, the model possesses efficient feature selection and complex feature interaction capabilities. Model performance was evaluated using 10-fold cross-validation. The proposed model achieves a CIE LAB color difference of 2.71 units and a CIE DE2000 color difference of 2.08 units on average, which outperforms the partial least squares regression, the radial basis function neural network, and so on. The proposed model can capture colorimetric characteristics of a color imaging system more effectively and enhance detail preservation. This research improves the accuracy of colorimetric characterization and can provide higher accuracy in colorimetric measurement for high dynamic range imaging.