Since digitization in the late 20th century, there has been a signicant advancement in how people interact with data, culture, and information systems, transforming these interactions signicantly. In academic research, systems utilizing digitized resources have promoted empirical research in various disciplines. The inclusion of machine and deep learning components in these systems has introduced novel approaches to research. This is particularly evident in the eld of Digital Humanities (DH), where innovative and diverse research methodologies are emerging, although these have primarily been centered around texts.

Lately, advancements in image processing and deep learning have shown promise in extracting information from images at large scales, suggesting a new direction for image-based information systems. This is key to overcoming the traditional limitations associated with image utilization in information systems. While some initiatives have attempted to integrate deep learning and image processing into these systems, there is a gap in understanding and addressing the varied needs and viewpoints of users, which is crucial for creating usable software. Understanding the unique research approaches and information requirements of DH scholars, known for their varied cognitive approaches to research, is deemed particularly essential.

This study investigates strategies for integrating computer vision, a subset of deep learning, to aid image-centric research within DH. It specically focuses on 19th-century children's and young adult literature illustrations, a subject of particular interest among DH researchers, to meet their special needs and viewpoints. This research incorporates the outcomes of various deep model applications into the Illustration Information System (IlluInfo), actively involving users throughout the development stages to ensure their diverse needs and perspectives are well represented and addressed in the system. The study not only explores the feasibility of applying deep learning to illustrations, often considered ambiguous and unreliable, but also assesses their acceptance in research settings. The _ndings of this research underscore the practical applications of deep learning in DH, highlight ways to involve users in the development cycle, and o_er valuable evaluations that could enhance system development and acceptance.