본문 바로가기
  • Home

An Inclusive AI Docent System for Accessible and Interactive Art Appreciation using Vision and Language Models

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(6), pp.109~118
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : April 24, 2025
  • Accepted : June 19, 2025
  • Published : June 30, 2025

Min-Su Kim 1 Min Kim 2 Hyeon-jung Kwak 3 Ye-jun Choi 4 Hyung-rok Lee 5 Chi-wook Ahn 6 Won Joo Lee 7 Young-Bok Cho 8

1대진대학교
2명지대학교
3이화여자대학교
4인하대학교
5연세대학교
6아주대학교
7인하공업전문대학
8국립경국대학교

Accredited

ABSTRACT

In this paper, we propose an interactive art appreciation system that integrates computer vision and large language models to enable users—including those with visual impairments—to actively engage with visual art. The system recognizes artworks using YOLO-based object detection and VGG16 classification, and applies HSV-based color correction to enhance the reliability of emotional analysis. Subsequently, the Qwen2.5-VL-3B model summarizes visual content, while the Qwen2.5-32B model generates emotionally enriched descriptions and facilitates interactive dialogues. Additionally, a retrieval-augmented generation (RAG) framework is implemented to answer user questions, and prompts inspired by Visual Thinking Strategies (VTS) are used to elicit emotional responses and foster meaningful engagement. The proposed system demonstrates the potential of artificial intelligence to enhance both the accessibility and immersiveness of art appreciation, offering a new model for inclusive interaction with visual culture.

Citation status

* References for papers published after 2023 are currently being built.