Multispectral imaging colorimetric measurement method with illumination-adaptive white calibration

Accurate and standardized color measurement of non-planar surfaces in non-contact scenarios remains a challenge, primarily due to complex measurement conditions such as uncontrolled illumination and variable surface scattering properties. To address this issue, a multispectral imaging colorimetric measurement method for non-planar surfaces based on uniform illumination and illumination-adaptive white calibration was proposed. The colorimetric characterization method relies on spectral reconstruction using a multiple-channel input backpropagation neural network, which processes the imaging system's output with illumination-adaptive white calibration and spectral fusion techniques. A device has been developed following this method, consisting of a large integrating sphere, an eight-channel multispectral camera, and a stepper motor-driven rotating table. To reduce the impact from the illumination, the rotating table, coated with polytetrafluoroethylene (PTFE), allows for sample placement and illumination-adaptive white calibration. After applying PTFE data correction to the test samples, the average reconstruction error decreased from 2.4 to 1.5 in terms of CIEDE2000 color difference units. To assess the measurement deviation on glossy curved surfaces at different angles, we tested it on a blue-and-white porcelain object with a smooth glaze. The ΔE₀₀ remained below 2 across observation angles, demonstrating superior accuracy over conventional point light sources for glossy objects. The device was subsequently used to measure porcelain exhibits at the Forbidden City. Experimental results confirmed that the proposed method ensures uniform and stable diffuse lighting and effectively combines digital imaging with spectroscopy, making it a powerful tool for precise color analysis in the non-contact domain.