On Color Cluster Analysis with Three-dimensional Fuzzy Color Ball

Dae‐Won KimJournal of Korean Institute of Intelligent Systems 2008 · DOI

Abstract

The focus of this paper is on devising an efficient clustering task for arbitrary color data. In order to tackle this problem, the inherent uncertainty and vagueness of color are represented by a fuzzy color model. By taking a fuzzy approach to color representation, the proposed model makes a soft decision for the vague regions between neighboring colors. A definition on a three-dimensional fuzzy color ball is introduced, and the degree of membership of color is computed by employing a distance measure between a fuzzy color and color data. With the fuzzy color model, a novel fuzzy clustering algorithm for efficient partition of color data is developed.

Introduction

Color perception is inherently subjective and influenced by environmental factors such as simultaneous contrast. Traditional color models (e.g., RGB, HSV) often treat color as "crisp" data points, failing to account for the ambiguous transitions between colors. This paper addresses two primary challenges in color analysis: perceptual non-uniformity—where different color regions have varying perceptual sizes—and the lack of a mathematical framework to describe uncertain boundaries between neighboring colors. By bridging mathematical representation with human psychophysics, the authors propose a model that better reflects human color perception.

The Fuzzy Color Ball Model

The core of the proposed approach is the Fuzzy Color Ball. Instead of representing a color as a single point, it is modeled as a three-dimensional ball within the perceptually uniform CIELAB color space. The radius of each ball is determined by the Just Noticeable Difference (JND), the threshold at which a human observer can distinguish between two colors.

Conceptual representation of fuzzy color balls and their overlapping membership functions.

A modified distance measure, $\delta(x, \tilde{c}_i)$, calculates the relationship between a color element $x$ and a fuzzy color ball $\tilde{c}_i$. If a color falls within the JND radius, its membership degree is 1.0. If it falls within the JND of a different color, its membership is 0.0. For colors in "vague" intermediate regions, membership is calculated based on relative distances, allowing for "soft" classification decisions.

Fuzzy Clustering Methodology

The authors extend the standard Fuzzy C-Means (FCM) algorithm by modifying the objective function to minimize the sum of squared distances between color elements and fuzzy color centroids (the balls).

To prevent the algorithm from converging to local minima—a common issue in clustering—the authors utilize a Munsell-based initialization. Ten reference colors from the Munsell color system (e.g., Red, Yellow-Red, Green) are used to identify dominant colors in the dataset and provide a perceptually grounded starting point for the cluster centroids.

Experimental Results

The algorithm was evaluated using a synthetic CIELAB dataset consisting of 15 color elements. The goal was to partition these elements into three distinct clusters ($p=3$).

Evolution of membership values for 15 color elements across three clusters over 15 iterations.

As shown in the iteration table, the algorithm reached a stable state in 15 iterations. The results demonstrate that colors clearly within a JND radius maintained high membership (near 1.0), while "vague" colors (such as $x_6$) saw their membership degrees evolve as the centroids shifted, eventually leading to a more certain classification. This highlights the algorithm's ability to handle uncertainty more effectively than traditional methods.

Conclusion

This research introduces a robust framework for color clustering by integrating psychophysical concepts like JND into a fuzzy logic model. While the current evaluation is limited to a small dataset and assumes spherical color balls, it provides a strong foundation for perceptually grounded color analysis. Future work will focus on scaling the algorithm for high-resolution images and exploring asymmetric color ball models to better match the complexities of human vision.

Summary updated

Abstract

The focus of this paper is on devising an efficient clustering task for arbitrary color data. In order to tackle this problem, the inherent uncertainty and vagueness of color are represented by a fuzzy color model. By taking a fuzzy approach to color representation, the proposed model makes a soft decision for the vague regions between neighboring colors. A definition on a three-dimensional fuzzy color ball is introduced, and the degree of membership of color is computed by employing a distance measure between a fuzzy color and color data. With the fuzzy color model, a novel fuzzy clustering algorithm for efficient partition of color data is developed.
BibTeX
@article{kim2008color,
  author = {Dae‐Won Kim},
  title = {{On Color Cluster Analysis with Three-dimensional Fuzzy Color Ball}},
  journal = {Journal of Korean Institute of Intelligent Systems},
  year = {2008},
  doi = {10.5391/jkiis.2008.18.2.262},
}
Figure