文化遺產圖像深度語意標引方法設計與實現

隨著文化遺產圖像資訊資源的迅速增長和數位人文研究的發展，針對文化遺產圖像語意內涵的深度語意標引逐漸引起人們的關注。文化遺產圖像的深度語意標引不僅可以提高圖像檢索與獲取的效率，增強使用者對圖像的理解程度，更能支援自動的圖像資源整合和知識發現，具有重要的理論和實踐意義。通過分析文化遺產圖像語意特徵及其主題、歸納總結現有文化遺產詮釋資料與本體組織，在明確深度語意標引的概念及其基本要求的基礎上，設計了深度語意標引的基本流程，構建了文化遺產圖像語意標引模型：宏觀概念模型、蘊含資訊層次模型與標引文本的結構化組織模型，並針對敦煌壁畫《九色鹿王本生圖》進行圖像深度語意標引實驗。圖像深度語意標引模型揭示了概念、圖像、文字等之間的語意關係，深度挖掘圖像標引中各資訊層之間的知識關聯，實現了圖像資訊單元細粒度組織。同時標引實驗驗證了圖像深度語意標引模型與標引文本結構化組織的可行性。文化遺產圖像深度語意標引方法的設計與實現是深度語意標引理論及圖像資訊組織理論的創新，對圖像資訊組織與數位人文研究發展具有深遠的意義。圖像深度語意標引的粒度與深度應以標注條件為基準進行選擇，深度語意標引資訊關聯集成與發佈展示也將成為未來研究發展的重點內容。

With the rapid growth of information resources for cultural heritage images and the development of Digital Humanities research, Deep Semantic Indexing (DSI), which aims at semantic indexing of cultural heritage images, has gradually attracted more and more attentions. DSI can improve not only the efficiency of image retrieval and acquisition, but also the user understanding of the images. It can support the integration of image resources and automatic knowledge discovery, which has important theoretical and practical significance. The study conducted throughout analysis of semantic features and themes of cultural heritage images and reviewed the existing cultural heritage metadata models and ontologies. Based on the understanding of the concept of DSI and its basic requirements, we designed the workflow and technological process of DSI, constructed the cultural heritage image semantic indexing model, including an inclusive concept model, an multi-layered information model, and a structural model of the indexing texts. We also conducted an indexing experiment of the Dunhuang mural “Nine-colored Deer”. The DSI modeling of images reveals the semantic relationship between concepts, images and text, mines the knowledge correlation between each information layer related to image indexing, and realizes the fine-grained organization of image information units. At the same time, the indexing experiment verified the feasibility and scientificity of cultural heritage image DSI structure. The design and implementation in DSI of cultural heritage image information is an advancement of the deep semantic indexing theory and image information organization theory. The decision on image indexing’s granularity and extensibility should be based on the indexing contents. The integration of DSI information and the publishing of such information will be studied further in the future.