Blue-and-white porcelain, as a distinguished representative of Chinese ceramic art, holds a prominent place in the global history of decorative arts. Its unique cobalt-based underglaze technique and iconic blue-and-white palette not only demonstrate exceptional craftsmanship but also embody profound cultural symbolism. The color characteristics of blue-and-white porcelain have evolved over time under the combined influence of raw material composition, glaze formulation, firing conditions and prevailing aesthetic values, forming distinctive historical trajectories. However, most existing studies have focused on qualitative explorations from aesthetic, semiotic and archaeological perspectives, with limited application of systematic quantitative methods, thus constraining objective comparisons and the identification of developmental patterns across different periods. With the rapid advancement of digital imaging, computer vision and machine learning, it has become possible to analyze artifact colors with precision using non-contact, repeatable and quantifiable techniques, offering new pathways for heritage preservation and revitalization. In this study, the K-means++ clustering algorithm was integrated with the Natural Color System (NCS) to digitally extract and parameterize the color features of blue-and-white porcelain from the Yuan, Ming and Qing dynasties, aiming to reveal the historical evolution of its hue, blackness and chromaticness, and hence establish a structured color database that can support heritage conservation, design innovation and cross-media application.

A total of 223 representative blue-and-white porcelain sherds from various historical periods were collected from the Tao Xi Chuan Ceramic Shard Museum in Jingdezhen and private collections. Compared with complete vessels, sherds offer flatter surfaces that minimize light distortion and facilitate accurate color capture, while their accessibility broadens the sample base for quantitative studies. Each sample was photographed in a controlled studio environment using a Canon EOS 6D camera at a resolution exceeding 300 dpi and 2000×2000 pixels. Lighting was maintained between 5000K and 6000K to simulate natural daylight, with diffusers and black light-absorbing cloth to eliminate direct glare, shadows and environmental reflections. A standard color card was included in each frame, while the images were color-corrected in Adobe Lightroom through white balance adjustment and matching to reference values, ensuring high fidelity between digital and physical color. The corrected images were processed using the K-means++ clustering algorithm. Unlike standard K-means, K-means++ optimizes the selection of initial centroids through probabilistic dispersion, thereby avoiding local minima and improving clustering stability. RGB values of each pixel were extracted, normalized and grouped into clusters, with the optimal cluster number determined by combining the elbow method, to identify the inflection point in the SSE curveand silhouette coefficient analysis, which measures cluster cohesion and separation. Both metrics consistently indicated K=5 as optimal for most periods. RGB values of the five dominant colors from each sample were then converted to the NCS parameters of hue (H), blackness (S) and chromaticness (C) for further statistical and comparative analysis.

The analysis revealed a consistent concentration of hues within the NCS range R90B–R60B across all three dynasties, indicating a strong bias toward cool blue tones with occasional shifts toward purpleblue. In the Yuan Dynasty, colors were characterized by high blackness (50–70) and low chromaticness (10–30), producing deep blue-gray tones. These visual attributes correspond with the use of both domestic and imported cobalt sources containing higher impurity levels, resulting in muted saturation. In the Ming Dynasty, particularly during the Yongle and Xuande reigns, the refinement of imported cobalt and improved firing techniques led to increased chromaticness (20–50, occasionally exceeding 60) and reduced blackness (30–60), producing purer brighter blues. Later Ming wares incorporated domestic cobalt, expanding the tonal range to include both light azure and intense deep blues. In the Qing Dynasty, especially during the Kangxi, Yongzheng and Qianlong reigns, colors exhibited greater hue diversity, with shifts toward R60B purpleblue tones. Blackness was moderate (30–50) and chromaticness frequently ranged from 40 to 70, producing vibrant saturated blues. These patterns confirm the correlation between technological refinement, material purity and aesthetic diversity, as well as the influence of court tastes and production scale on color variability.

It is demonstrated that integrating K-means++ clustering with the NCS system provides a precise, replicable and non-invasive approach for quantifying the color characteristics of ceramic artifacts. By revealing the historical trajectory of blue-and-white porcelain, from the Yuan Dynasty's deep, restrained palette, through the Ming Dynasty's balanced and refined blues, to the Qing Dynasty's diverse and saturated tones, a robust color reference framework was established, which is applicable to digital heritage archiving, cultural product development and cross-media design translation. Furthermore, the workflow presented here addresses a broader methodological gap in heritage studies, by bridging computational analysis with perceptual color systems. This enables not only the preservation of color data in standardized formats but also its transformation into actionable design strategies. Such strategies can inform visual identity systems, spatial design schemes and product color planning, ensuring that the cultural essence of traditional palettes is retained, while adapting to contemporary design languages. Ultimately, this approach advances both the academic study and practical application of historical color knowledge, fostering new avenues for innovation in the digital humanities and creative industries.