TLDR: The COLIBRI Fuzzy Model introduces a human perception-based color model that uses fuzzy sets and logic to bridge the gap between computational color representations and human visual perception. Unlike traditional rigid color models, COLIBRI accounts for the gradual, context-dependent, and linguistically influenced nature of human color perception. Developed through extensive human categorization experiments (n=2496), the model defines soft transitions between color categories, offering a more human-aligned classification. It has potential applications in AI, design, and image processing, particularly for tasks requiring perceptually relevant color representation.
Colors are everywhere, influencing our emotions, decisions, and how we interact with the world. Yet, for computers, truly understanding and replicating human color perception has been a significant hurdle. Traditional color models like RGB, HSV, and LAB, while widely used, struggle to capture the subtle, context-dependent, and often ambiguous nature of how humans see colors. These models rely on rigid boundaries, which don’t align with our fluid perception of color transitions.
Consider the flag of Kazakhstan. Officially described as ‘sky blue,’ it can appear cyan, turquoise, or even light blue depending on the image, display settings, lighting, or cultural context. This isn’t just a technical issue; it highlights a fundamental challenge in how we linguistically categorize colors. For instance, Russian and Kazakh have distinct terms for dark blue (‘siniy’) and light blue (‘goluboy’), whereas English uses a single term ‘blue’ for both. Such linguistic differences influence how people perceive and classify colors, creating ‘fuzzy’ boundaries in human vision. This inconsistency reveals the subjective nature of color perception.
To address this gap, researchers have introduced a new fuzzy-based color model called COLIBRI (Color Linguistic-Based Representation and Interpretation). This innovative model is designed to bridge the divide between computational color representations and human visual perception. It adapts to human perception and language-driven color distinctions, allowing for soft transitions between linguistic categories and offering a more human-aligned color classification. The full research paper can be found here: COLIBRI Fuzzy Model: Color Linguistic-Based Representation and Interpretation.
A New Approach to Color Modeling
COLIBRI is fundamentally linguistic, built on soft color categorization that mirrors how humans naturally perceive and name colors. Unlike rigid numerical models, COLIBRI uses ‘fuzzy sets’ to create smooth transitions between color categories, accommodating overlapping color terms. This model is particularly inclusive, even providing linguistic descriptions of colors for individuals with color blindness.
The development of COLIBRI involved a comprehensive, three-phase experimental approach with a large-scale human categorization survey involving over 1000 subjects. This extensive data was used to extract fuzzy partitions and generate ‘membership functions’ that reflect real-world perceptual uncertainty. The model also incorporates an adaptive mechanism, allowing it to refine its categorizations based on feedback and contextual changes.
Understanding How We See Color: The Experiments
The study’s methodology began with a multistage experimental framework to collect data on human color perception. This involved:
- Experiment 1: Perceptual Color Boundaries and Naming: Seven color experts identified and categorized distinct hue, saturation, and intensity groups. This experiment revealed nine linguistic hue categories, including ‘Light Blue’ as a distinct category, acknowledging cultural nuances in color naming.
- Experiment 2: Hue Stimuli Selection: A preliminary survey with 27 participants helped refine the specific color stimuli (45 hues) to be used in the main experiment, ensuring they were perceptually meaningful.
- Experiment 3 (Main): Human Color Stimuli Categorization: This was the core experiment, involving 1,071 participants for hue categorization and hundreds more for saturation and intensity. Participants classified color stimuli based on hue, saturation, and intensity. The hue categorization was conducted in two versions: a ‘main’ version requiring a single choice and an ‘alternative’ version allowing multiple selections, providing flexibility in how participants labeled colors.
The experiments were conducted in controlled university computer rooms with standardized settings, including lighting and viewing distance, and all instructions were provided in Kazakh, Russian, and English to accommodate diverse linguistic backgrounds.
Key Findings and Insights
The results from the main hue experiment showed a strong consensus among participants for many stimuli, with over 95% agreement on clear colors like red, green, and blue. However, some stimuli, particularly those near category boundaries (e.g., between cyan and light blue), showed moderate to substantial ambiguity, with responses nearly equally divided. This supports the idea that human color perception is not always clear-cut.
Interestingly, the analysis of hue categorization by gender showed minimal differences, suggesting a shared conceptual understanding of color categories between male and female participants within the study’s context. However, the Ishihara color blindness test, administered to participants, confirmed that males are more likely to be colorblind, aligning with existing research.
The study also found that certain hues, like green, blue, and magenta, have broader ‘fuzzy sets,’ indicating a wider range of human tolerance and consistent recognition. In contrast, colors like yellow, cyan, and light blue have narrower fuzzy sets, reflecting more distinct or ambiguous perception.
Also Read:
- A Unified Framework for Understanding Neurosymbolic AI
- Exploring the Creative Frontier: Can Artificial Intelligence Truly Be Creative?
Applications and Future Directions
The COLIBRI model has significant implications for various fields. In computer vision and AI, it can improve color selection, object segmentation, and content-based image retrieval by assigning linguistic labels that closely resemble human perception. For design and aesthetics, COLIBRI can help predict color harmony, benefiting branding, fashion, and architecture. In medical imaging, it could aid in diagnosing diseases by identifying subtle color variations in tissues.
The research acknowledges limitations, such as the influence of individual and cultural differences on color perception, and the model’s current inability to account for context-dependent color perception. Future work aims to incorporate more psychological aspects, like the emotional impact of color, and adapt the model for specific fields like fashion and architecture. This study provides a robust and reproducible framework for developing human-centered color models and perceptually driven AI systems.
