Disruptive Transformation of Artworks in Master-Disciple Relationships: The Case of Ukiyo-e Artworks

Honna Shinichi, Akira Matsui · 2025

Read Disruptive Transformation of Artworks in Master-Disciple Relationships: The Case of Ukiyo-e Artworks by Honna Shinichi, Akira Matsui (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-08284.txt.

Introduction

Honna Shinichi and Akira Matsui measure creativity in Japanese woodblock prints with network methods. Their paper appeared in two thousand twenty-five. Art history has long leaned on expert taste and judgment. Machine learning now lets scholars score traits no eye could tally. Western painting has seen such studies at scale. Eastern art has not, so the authors take Ukiyo-e as their case.

They analyze eleven thousand high-resolution images of prints. A network-based creativity score grades both works and artists. Links between masters and disciples frame the study. Across Ukiyo-e as a whole, freshness of appearance faded as the culture matured. Style tells a different story, which the paper pursues. The work shows how data methods can test claims about artistic change.

Catalog record

CREATORHonna Shinichi, Akira Matsui
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-08284

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Disruptive Transformation of Artworks in Master-Disciple Relationships: The Case of Ukiyo-e Artworks Honna Shinichi, Akira Matsui Abstract Artwork research has long relied on human sensibility and subjective judgment, but recent developments in machine learning have enabled the quantitative assessment of features that humans could not discover. In Western paintings, comprehensive analyses have been conducted from various perspectives in conjunction with large databases, but such extensive analysis has not been sufficiently conducted for Eastern paintings. Then, we focus on Ukiyo-e, a traditional Japanese art form, as a case study of Eastern paintings, and conduct a quantitative analysis of creativity in works of art using 11,000 high-resolution images. This involves using the concept of calculating creativity from networks to analyze both the creativity of the artwork and that of the artists. As a result, In terms of Ukiyo-e as a whole, it was found that the creativity of its appearance has declined with the maturation of culture, but in terms of style, it has become more segmented with the maturation of culture and has maintained a high level of creativity. This not only provides new insights into the study of Ukiyo-e but also shows how Ukiyo-e has evolved within the ongoing cultural history, playing a culturally significant role in the analysis of Eastern art. 1 Introduction Art serves significant cultural and historical roles in our society and artists mutually influence each other [ 1 ] . While recent research reveals connections among contemporary artists [ 1 , 2 ] , we understand little about the mutual relationships among historical artists in their creation of artworks in a quantitative way. Art historians qualitatively analyze and interpret the value and meaning of artworks based on style, theme, technique, and cultural context, leading to subjective evaluations influenced by their professional experiences [ 3 ] . Most research relies on historians’ tacit knowledge, focusing mainly on individual works or specific artists rather than the history of art as a whole [ 4 ] . While evaluations and research by art historians significantly contribute to our knowledge, the reliance on qualitative measurements results in a lack of longitudinal analysis from massive historical records. Recently, researchers have been conducting qualitative evaluation of art [ 5 ] using machine learning models powered by digital technologies [ 6 , 7 ] , assessing creativity from images of artworks [ 8 , 9 , 10 , 11 ] . For example, Convolutional Neural Network (CNN) can identify unique brushstrokes of different painters attribution [ 12 ] , and image recognition techniques are also popular [ 13 , 14 , 15 ] . Such studies learn the characteristics of the works for applications in art historical research, cataloging, and even detecting forgeries [ 16 , 17 ] . In other qualitative research on artworks with machine learning, researchers can detect compositional patterns that serve as artist or style signatures and reveal non-obvious historical groupings through network analysis [ 18 ] . Not only CNN, but also an advanced multimodal AI (Stable Diffusion), can analyze 500 years of Western painting and reveal the importance of contextual information over formal elements retrieved from painting images [ 19 ] . Those advancement of technology and data availability enables us to conduct large-scale quantitative analysis, for example using image recognition techniques [ 13 , 14 , 15 ] . Thanks to such unsupervised models we uncover the hidden structure or mechanism from the artwork in a data-drive manner [ 9 , 20 ] , but the evolution of artwork and artists’ mutual interdependence lacks transparency. There is, for instance, a universal pattern of artists’ performance within their career trajectory [ 20 ] . Additionally, we know little about how mutual influence drives creativity. While we understand the impact of mutual influence on creativity in scientific teams [ 21 , 22 ] , this understanding does not fully translate to the realm of artistic creativity. While the connections between paintings can estimate creativity of artworks [ 8 ] and their diversity patterns linked to emerging styles [ 23 ] , we need to have a model that tract the trajectory of artists creativity extracted from their connections thorough artworks. To address the aforementioned gap in the quantitative analysis of artworks, we propose a machine learning-based approach to studying the influence of an artist’s master-disciple relationship on their work style. We construct an artist network that captures the extent to which created artworks are influenced by past works, and we calculate the trajectory of disruption [ 24 ] over 100 years to study their creativity. We apply the proposed framework to the historical Japanese Ukiyo-e artwork dataset, enabling the analysis of long-term artist influences. The large number of widely appreciated artworks, the period of cultural isolation, and the strong master-disciple system provide a clear basis for examining mutual artistic influences [ 25 , 26 , 27 ] . Figure 1 : Conceptual Diagram of the Model Used (a) The figure schematically shows a fine-tuned VGG model. The part with frozen weights has fixed weights, while the part with trainable weights is being trained. The model classifies ukiyo-e images by determining which ukiyo-e artist created them when we input an ukiyo-e image. (b) The features obtained from the layers within the model are displayed as a heatmap. High-dimensional features extract linear characteristics, while low-dimensional features capture the differences in painting styles specific to each artist. 2 Data and Methods 2.1 Dataset This research learns the embedding of Ukiyo-e artwork from the “ARC Ukiyo-e Face Dataset“ [ 28 ] . This dataset contains information such as ID, title, production era, author name, and images of Ukiyo-e for each artwork. The central aim of the dataset is to provide portraits of people and therefore does not include landscape paintings. It contains about 11,000 images and corresponding metadata. Among the painters in this dataset, those who have more than 100 Ukiyo-e prints in the dataset are Kunisada, Kunisada, Kunisada, Toyokuni, Toyokuni, Kuniyoshi, Kogyo, Kunichika, and Hirosada (Table 1 ). We plot the sample of those artists from the dataset in Figure 2 . These Ukiyo-e artists have more than 100 pictures in the dataset that have all three pieces of information (image, artist name, and created year). In this study, we obtain the characteristics of the painting style from the deep learning model by classifying the representative Ukiyo-e artists who painted many Ukiyo-e works. The reason for using this method is that there is no annotation data for each painter. In Western painting, there are painting styles such as Impressionism, but in Ukiyo-e, painting styles are not explicitly assigned to each Ukiyo-e artist. Figure 3 indicates the age of each artist’s activity. For each artist, a box-and-whisker chart was used to visualize which period of roughly 100 years the artist was active, from Toyokuni I to Kogyo. And, Figure 11 shows that the most Ukiyo-e artists were active between 1850 and 1860, and the number of Ukiyo-e works during this period is the largest for Ukiyo-e as a whole. Table 1: Summary of Ukiyo-e Artists and Image Counts ID Ukiyo-e Artist (Active Period) Number of Images Used Generation (if applicable) 1 Kunisada (1808–1843) 245 1st 2 Kunisada (1851–1870) 146 2nd 3 Kunisada

Read the complete text