一個台灣手語詞典檢索的參考系統架構

基於認知心理的雙編碼理論，我們將每個手語字詞的文字與視覺兩部份分別建立索引，期望經由手勢視覺特徵檢索可以擷取每個字詞。然而，近年來的台灣手語線上詞典的研究，其手勢的視覺特徵檢索功能僅能透過單隻手的手型與位置，目前尚未整合手語社群所認可的完整視覺特徵。以至於當人們以手語溝通時，遭遇到不認識的手語生字，卻無法以所有可能的視覺特徵來檢索詞典。因此，本研究假定一個手語字詞包含多個手勢數位圖像，而每個圖像由多個視覺特徵所組成的序列來標記。我們採用一個壓縮率較高且檔案較小的向量式可縮放向量圖形（SVG）來表示每個手勢圖像，可節省網路傳送成本。由於目前台灣手語社群的此種檢索需求不明確，如何建構一個社群所認可的詞典檢索系統架構，作為系統開發之參考藍圖，來整合社群的需求，便顯得其重要性。因此本研究目的乃依據手語社群所認可的完整視覺特徵，提出一個台灣手語詞典檢索參考系統架構（Reference System Architecture for Taiwanese Sign Language Dictionary Retrieval, RSA4TSLDR），它包含文字與視覺特徵檢索的架構上下文、應用架構及資料架構等描述。並基於此參考架構實際開發一個台灣手語詞典檢索雛形系統（Taiwanese Sign Language Dictionary Retrieval Prototype System, TSLDRPS），以作為手語詞典編輯者及詞典檢索者的平台。最後，我們針對RSA4TSLDR與TSLDRPS進行德菲法（Delphi）專家問卷調查，並根據調查的結果去修正兩者。

關鍵字

圖像檢索；台灣手語；圖像元搜索；圖像標記；系統架構

並列摘要

Based on the dual coding theory in cognitive psychology, we expect to retrieve each sign language word by indexing its text and visual gesture features respectively. However, in recent research on Taiwanese Sign Language online dictionaries, the visual feature query function of gestures can only retrieve target words by the hand shape and the position of a single hand and has not yet integrated the complete visual features recognized by the sign language community for retrieval. So that on facing a communication problem when people encounter unknown words in sign language, they cannot retrieve the dictionary through all possible visual features. Therefore, this study assumes that a sign language word contains multiple digital images of gestures, and each image is tagged by a sequence of multiple visual features. We apply an SVG (Scalable Vector Graphic) image with a small-sized and well-compressed file to represent the gesture image could save space and transmission time. Since the current retrieval requirements of the Taiwanese sign language community are not clear, how to construct a system architecture recognized by the community for such retrieval as a reference blueprint for system development to integrate the requirements of the community is important. Therefore, the aim of this study is to propose a Reference System Architecture for Taiwanese Sign Language Dictionary Retrieval (RSA4TSLDR) based on the complete visual features recognized by the sign language community. RSA4TSLDR contains the descriptions of the architectural context, application architecture and data architecture about the text and visual feature retrieval. Based on the architecture, a Taiwanese Sign Language Dictionary Retrieval Prototype System (TSLDRPS) is developed to serve as a platform for editors and retrievers of the dictionary. Finally, we have conducted a Delphi expert questionnaire on RSA4TSLDR and TSLDRPS and have evaluated and revised them according to the results of the survey.

並列關鍵字

Image Retrieval ； Taiwanese Sign Language ； Image Meta Search ； Image Tagging ； System Architecture

主題瀏覽