Comparative Analysis of Clustering Algorithms for Human-Consistent Dominant Color Extraction

Capturing human-perceived colors in digital images remains a significant challenge, as traditional clustering methods often fail to reflect human visual perception. This study presents a comparative analysis of clustering algorithms for human-consistent dominant color extraction. Traditional methods, such as RGB histograms and k-means clustering, are compared with more advanced techniques, including fuzzy clustering, mean shift clustering, and Gaussian mixture models. The comparative evaluation experiment assesses the methods for aligning with human perception. To assess the perceptual accuracy of these methods, an experimental study involving 100 human participants is conducted. Using 20 image stimuli, participants evaluate the extracted colors from each algorithm and select those they perceive as most representative of the original image. The study findings show how different clustering approaches capture color saliency and consistency from a human perspective. Our results show that the hierarchical clustering algorithm achieved the highest selection rate, indicating the closest alignment with human perception. Findings from this study contribute to the development of human-centric color modeling, improving applications in image processing, design, and digital media.