Computational color prediction versus least-dissimilar matching.

Area of Science:

• Color Science
• Visual Perception
• Computational Imaging

Background:

• Asymmetric color matching experiments investigate how observers perceive color differences under varying illuminants.
• Evaluating the accuracy of computational color prediction models against human perception is crucial for developing robust color appearance models.

Purpose of the Study:

• To compare the performance of various color prediction methods against human observer data in an asymmetric color matching task.
• To investigate whether individual observers or computational models better predict average observer matches.

Main Methods:

• Six color prediction methods (CIECAM02, KSM2, Waypoint, Best Linear, Metamer Mismatch Volume Center, Relit) were evaluated.
• Human observers performed asymmetric color matching, identifying the least-dissimilar Munsell paper under a test illuminant.
• A leave-one-observer-out analysis was used to compare individual observer performance against computational models.

Main Results:

• The mean color signal of matched papers closely approximated the color signal of the physically identical paper under the match illuminant.
• Surprisingly, individual observers' average matches were predicted more accurately by other individual observers than by any of the tested computational color prediction methods.

Conclusions:

• Human visual perception demonstrates a high degree of consistency in asymmetric color matching tasks.
• Current computational color prediction models may not fully capture the nuances of human color perception compared to inter-observer agreement.