Digital Humanities: a bridge between computer vision and study of art

From Humanities Computing to Digital Humanities, the development of digital technologies has brought many possibilities to the Humanities disciplines, including the exploration of painting research through deep learning. Such research currently focuses primarily on improving algorithmic performance, mainly derived from the field of computer vision, while technical barriers and disciplinary jargon make it difficult for Humanities scholars to engage in this type of research. However, effective interdisciplinary research requires communication and dialogue across multiple fields. On the one hand, the participation of Humanities scholars can make deep learning technologies more targeted in exploring painting research, thereby providing valuable research insights. On the other hand, Humanities scholars can critically examine deep learning, offering feasible suggestions for technological improvement, and identifying and avoiding potential ethical issues. Based on this research objective, this thesis conducts the following studies. The thesis first introduces the context of this research: “Digital Humanities,” “Artificial Intelligence,” “Computer Vision,” “Digital Art History,” and “Cultural Analytics.” and reviews existing research on deep learning in painting studies, critical research in Digital Humanities, and ethical issues in AI. Then, the thesis details the application process of deep learning in painting research in four parts: “Data Preparation,” “Model Training,” “Evaluation and Optimization,” and “Analysis and Interpretation,” each part including an introduction to basic knowledge, the application of technology (experiments), and reflections on deep learning. Chapter One, Data Preparation, introduces the basics of art image datasets, discussing how to assess, select, and clean image datasets. The experiment demonstrates how to organize datasets with code according to one’s research objectives, preparing for model training. The reflective section discusses the subjectivity and biases of datasets, the characteristics of art data itself, and the ethical, copyright, and technical limitations of datasets, proposing some targeted and feasible suggestions. Chapter Two, Model Training, uses CNNs as examples to introduce the internal structure of neural networks and various types of CNNs. The experiment demonstrates how to train a simple neural network model to predict the authors of paintings. Lastly, it reflects from a Humanities perspective on the potential issues that may arise during the model training process, including programming challenges, ethical issues and transparency of algorithms, the comparison between machine learning and human learning, and the effectiveness of transfer learning in art images. Chapter Three, Evaluation and Optimization, introduces methods for evaluating and optimizing models, and through experiments, evaluates and optimizes the model trained in Chapter Two. The reflection section discusses minimal computation, model accuracy, as well as issues with weights and biases. Chapter Four, Analysis and Interpretation, differentiates between model explanations from a computational perspective and model interpretations from a Humanities perspective. Based on the objectives and themes of painting research, it proposes five potential ways in which deep learning can assist painting research, including recognition of features and patterns, comparative studies, object detection, unsupervised clustering, and image generation, with detailed case explanations for each path. After completing the introduction and exploration of deep learning, this chapter ultimately suggests a research approach that combines Humanities and computing and discusses the value of AI-generated art based on current research trends. The biggest contribution of this thesis is in linking the fields of painting research and deep learning, which have vastly different research goals and methods, to form a research approach that allows for mutual communication and contribution. This approach represents a Digital Humanities research approach, promotes equal and dialogical exploration between Humanities research and Computer Science, where technology offers new insights and possibilities for Humanities research, and Humanities research provides critical suggestions for technological development.

Grantor dc:publisher

University College Cork

Year dc:date.issued

2024

Author dc:creator

• Xiao, Shuang

Advisors dc:contributor.advisor

• Murphy, Orla
• Allen, Graham

• Digital Humanities
• Computer vision
• Study of art
• Digital art history
• Deep learning

Statement dc:rights

• © 2024, Shuang Xiao.

Licence dc:rights.uri

• https://creativecommons.org/licenses/by-nc-nd/4.0/

Language dc:language.iso

en

Handle dc:identifier.uri

https://hdl.handle.net/10468/16500

OAI identifier oai:identifier

oai:cora.ucc.ie:10468/16500