Recent advances in artificial intelligence (AI)–based image generation have raised questions about how colors of natural objects are reproduced. While AI systems generate colors based on statistical patterns learned from large-scale image data, human color perception is organized around categorical structures shaped by linguistic and perceptual experience. In particular, natural object colors are closely associated with human color naming categories, which are characterized by prototypical centers and fuzzy boundaries rather than simple statistical averages. This study reinterprets AI color reproduction of natural objects through the framework of human color naming categories established in prior experimental research. Drawing on findings from the previous study that empirically identified the distributional structure of human color categories, this research adopts human color categories as a perceptual reference framework. In addition, insights from recent research on AI-based color naming are incorporated to account for the instability of AI responses in category boundary regions. By analyzing the correspondence between natural object colors and human color categories, this study identifies characteristic patterns of category alignment, boundary proximity, and category deviation in AI color reproduction. Particular attention is given to the effects of statistical averaging, which can lead to chroma reduction and weakened color identity when natural object color distributions are summarized without regard to categorical structure. Based on these observations, the study introduces the concept of perceptual color reproduction fairness, defined as the degree to which reproduced colors maintain perceptual consistency within human color categories. Rather than equating fairness with statistical accuracy or average color values, this approach emphasizes the importance of preserving prototypical category-based color identity. The findings suggest that incorporating human color category structure provides a meaningful framework for evaluating and interpreting AI color reproduction of natural objects.
Han et al. (Sat,) studied this question.