The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Share
Publications Copernicus
Download
Citation
Share
Articles | Volume XLVIII-M-9-2025
https://doi.org/10.5194/isprs-archives-XLVIII-M-9-2025-1395-2025
https://doi.org/10.5194/isprs-archives-XLVIII-M-9-2025-1395-2025
03 Oct 2025
 | 03 Oct 2025

Study on the Color Characteristics of Reproduced Oil Paintings Using a Machine Learning Algorithm

Hyeong Rok Song and Young Hoon Jo

Keywords: Color Characteristics, Digital Color Reproduction, Machine Learning, Pixel-based Statistics, Oil Painting

Abstract. Accurate recording of the colors in cultural heritage is essential, where color data is crucial for various applications, including conservation, restoration, research, analysis, and archiving. Recently, advancements in digital color reproduction techniques have emerged, enabling precise documentation of colors using digital photography and image processing. This approach reproduces colors that closely match those of the cultural heritage object by correcting the image and profiling the camera. Notably, the correction process utilizes the device-independent CIE L*a*b* color space to ensure that the reproduced colors are consistent across different devices. Moreover, digital images consist of pixels, which facilitate data-driven statistical analysis. This study focused on digital color reproduction for Korean modern oil paintings, following a systematic process that included photography, digital color correction, and digital color space configuration. To enhance the reliability of color reproduction, it compared the spectral color measurement results of a color chart with the color differences observed in the reproduced images. The study then plotted the CIE L*a*b* color distribution of the images in a three-dimensional graph, where approximately 30 million pixels were classified using the K-means machine learning algorithm. Based on these classification results, representative colors were extracted, along with various analytical outcomes, such as the number of pixels, representative CIE L*a*b* color coordinates, and the percentage composition of each representative color. This research enabled oil paintings to be documented with accurate colors, and the resultant image data were used to extract representative colors using a machine learning algorithm. This method, wherein representative colors are derived through color reproduction, offers valuable insights into the color usage patterns and chromatic painting techniques of an artist, and even the authenticity of artworks.

Share