Evaluation and Adaption of Color Calibration Techniques for Non-uniformly Distributed Reference Colors

Color calibration, or more broadly, color management, is a fundamental component of any digitization or printing workflow, ensuring consistent and accurate color reproduction across different devices. Typically, color calibration involves capturing an image of a known color target, comparing the recorded colors to their corresponding reference values (usually measured with a spectrophotometer), and applying a calibration algorithm to compute a transformation that corrects the colors across the entire image. During the development of a facial scanner, we encountered issues with the color accuracy of the captured images, specifically regarding the reproduction of human skin tones. To address this problem, we opted to use a non-proprietary, selfprinted color target designed to optimize calibration results for skin color representation. This raised several research questions: How many color patches are required for effective calibration? How does the distribution of reference colors on a target influence calibration accuracy? And what are the effects of incorporating use-case-specific colors into a target? In this Bachelor’s thesis, we present a digital testing pipeline for evaluating color calibration methods and color targets using synthetic test charts and images. The process involves generating digital color targets, applying controlled color distortions to simulate calibration errors, and using a variety of calibration algorithms to correct the distorted images. This pipeline enables the design and evaluation of color targets for both general and application-specific use cases, without the need to physically print and test each variation. Using this pipeline, along with real-world validation, we demonstrate that most color calibration algorithms benefit from an increased number of color patches. We examine various mathematically derived color distributions and find that a distribution based on the HSL color space significantly improves calibration performance when compared to the MacBeth ColorChecker [16] and the RGB cube-based approach proposed by Barbero et al. [1]. Additionally, we investigate the impact of incorporating use-case-specific colors—such as skin tones—into
calibration targets and discuss the associated benefits and challenges. Finally, we validate the results obtained from our digital pipeline in three real-world setups using self-printed color targets. By analyzing their print quality and calibration performance in comparison to larger test targets, we confirm the consistency of our digital findings and demonstrate the effectiveness of using a digital design process as a viable approach for developing optimized color calibration targets.