Driven by digital governance and open interoperability, the digitization of image-based cultural heritage is being shifted from “collection–display” toward a closed-loop workflow of “semantic specification–controlled generation–online pres...

http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
https://www.riss.kr/link?id=T17393164
대구 : 경북대학교 대학원, 2025
2025
한국어
대구
viii, 289 p. ; 26 cm
지도교수: 이경용
I804:22001-000000112955
0
상세조회0
다운로드다국어 초록 (Multilingual Abstract)
Driven by digital governance and open interoperability, the digitization of image-based cultural heritage is being shifted from “collection–display” toward a closed-loop workflow of “semantic specification–controlled generation–online pres...
Driven by digital governance and open interoperability, the digitization of image-based cultural heritage is being shifted from “collection–display” toward a closed-loop workflow of “semantic specification–controlled generation–online presentation and evaluation.” Ming Dynasty blue-and-white porcelain—whose stylistic features are highly standardized and readily formalized—was selected as the study object, and the deployment of generative AI was examined for controllability, interpretability, and auditability across the “preservation–creation –transmission” chain, thereby addressing inconsistent semantic definitions and the lack of an end-to-end evidence chain linking data provenance, generation processes, and user evaluations.
A technical pipeline of “Common Expression Element System (CEES)–controlled generation–online presentation and evaluation” was constructed and implemented. Controlled vocabularies, value ranges, and compliance constraints on allowable combinations were defined for vessel form, compositional templates, motifs and borders, linework, and color/shading, thereby yielding a reusable minimal-viable combination set and an annotation schema. Semi-structured prompt templates, a style lock, and a negative list were embedded in a cross-platform Web/App prototype, while prompts and parameters were rule-validated and, when necessary, rolled back via a Style Control Card (SCC), so that controllable outputs could be produced when text-to-image generation was invoked using GPT–DALL·E. Prompts, negative terms, selected forms/patterns/parameters, reference and provenance information, and sketch inputs were captured in a structured, field-level generation log, thereby supporting reproducibility, traceability, and auditing.
For evaluation, evidence was collected at two levels (static viewing and mechanism analysis), and validation was conducted using Stylistic Fidelity (SF), perceived authenticity (PA), User Engagement Scale (UES), the Unified Theory of Acceptance and Use of Technology (UTAUT) scale, Cognitive Load Theory (CLT), and Torrance Tests of Creative Thinking (TTCT); additionally, the adoption mechanism was tested via an integrated PLS-SEM→ANN→NCA framework. The results indicate that, under controlled generation, SF and PA were maintained at an overall high level. Superior performance was observed for experts on style-sensitive dimensions and on Constructive Authenticity (CA) within PA, whereas non-experts exhibited greater novelty, originality, and Willingness To Use (WTU) and scored higher on Object Authenticity (OA) and Existential Authenticity – Self (EA-S) within PA. Overall differences in Extraneous Cognitive Load (ECL) were small. Mechanism analyses supported a path in which Authenticity of Cultural Elements (ACE) and Perceived Personalization (PP) affected WTU via Cultural Identity (CI), Perceived Usefulness (PU), and Perceived Ease of Use (PEOU); ANN and NCA further corroborated nonlinear importance patterns of key antecedents and identified necessary conditions and threshold constraints.
The proposed closed loop—“semantic specification–controlled generation– log-based provenance–online evaluation”—offers a transferable, reusable, and auditable generative-AI paradigm for image-based cultural heritage by operationalizing preservation norms as generation constraints, strengthening governance and reproducibility through logging and provenance mechanisms, and providing comparable empirical evidence for the modern transmission of Ming Dynasty blue-and-white porcelain visual vocabulary.
Keywords:Ming Dynasty blue-and-white porcelain; Common Expression Element System (CEES); controlled generation; Stylistic Fidelity (SF); perceived authenticity (PA); User Engagement Scale (UES); Unified Theory of Acceptance and Use of Technology (UTAUT) scale; Cognitive Load Theory (CLT); Torrance Tests of Creative Thinking (TTCT); Partial Least Squares Structural Equation Modeling (PLS-SEM); Artificial Neural Network (ANN); Necessary Condition Analysis (NCA).
목차 (Table of Contents)