發展情境描述模型應用於圖片檢索

A Scenario Description Model for Image Retrieval

蕭金美(Chin-Mei Hsiao)

指導教授 ： 戚玉樑

中原大學/商學院/資訊管理研究所/碩士(2010年)

https://doi.org/10.6840/cycu201000914

摘要

隨著電子科技的進步，使得數位圖片在拍攝、儲存和處理上更加簡單及便利，然而大量產生的圖片，卻也造成檢索及分類管理上的困難。圖片並不是一段段文字性描述的檔案，而是包含視覺上人物角色、表情、動作及周遭環境等，相互交織所營造的影像資料，由於文字比對的效果不佳，有些圖片檢索系統會建立詮釋記錄，如：題名、作者、建檔日期及格式等做為索引。這些欄位式的註記資訊主要針對圖片的外觀特徵進行描述，多半缺乏圖片內容詳細的著墨；此外，註記人員和檢索者對於圖片內容認知上的差異，常造成檢索者無法順利使用關鍵字找出正確的結果。為協助檢索者能夠查找心目中所預期的圖片內容，本研究提出以情境故事法中，常用於分析情境的5W概念（Who, When, What, Where, Why），做為描述圖片情境的方法並結合本體，發展以5W為基礎的人物圖片知識本體（ontology）。不同於扁平的資料庫資料，知識本體具備領域知識的階層架構，可賦予知識元素關聯或繼承等關係，使知識具有運算的能力。同時，為使圖片註記及檢索上有一個統一的用詞標準，本研究改良索引典（Thesaurus）蒐集詞彙的方法，蒐集及篩選可統一詮釋及代表人物圖片的知識素材，歸屬至5W的五大分類架構下，讓註記人員可以依照5W的描述架構，有系統及連貫性地根據圖片的內容物，循序為圖中的影像物件進行註記；同樣地，檢索者可將心目中想要的圖片畫面，透過5W架構發展圖片內容所需求的5W影像，進而篩選出符合情境的圖片。經由本研究系統評估的實驗結果證明，情境故事法與知識本體的結合，確實可協助使用者透過描述故事的情境勾勒出心中希望的圖片內容，大幅度地提高檢索圖片之精確性。

關鍵字

情境分析 ； 圖片內容檢索 ； 本體論 ； 情境故事法

並列摘要

With electronic technology advancement, taking, storing or processing digital photos will be easier and more convenient. However, a large number of pictures also cause a lot of problems for retrieving and category management. A picture is unlike a text file, which is described by contexts. It is an interlaced scenario that consists of a role, facial expression and action of a person from the vision and circumstance. Because the search result is not effective in text-compared manner, some picture search systems would make an extra Metadata, such as subjects, authors, building date, and even format to help with indexing. For the most part, the annotation information of columns is described from the feature of picture appearance, so lacked the detailed contents. In addition, within the information content of a picture, the different recognition between annotators and searchers will also make searchers can not use keywords to get satisfactory result. To assist searchers to sift the expected one out lots of pictures, this research proposes that using scenario , which is a common method to analyze scenario in 5W concept（Who, When, What, Where, Why）, is an alternative to describe picture and combines this with ontology, to develop the 5W as the based ontology of a person picture. Different from traditional flat database data, ontology covers relative or inherit hierarchy of each knowledge element so that it is able to perform the operation. Moreover, in order to provide a phrase standard for annotating and searching pictures, the research reforms, ‘Thesaurus’, a collect methods, to collect and sieve out knowledge materials, which can express and represent picture consistently, and then to group under five main categories further. Annotators can systematically and constantly follow 5W description architecture to make a annotation according to each image from the picture contents. In the same way, searchers also can put expected content of the picture by 5W framework to develop desire 5W images, through joining these images to get nearly picture which corresponded with scenario. After system evaluation in this research, combining scenario story and ontology can help users to draw out the desire picture through scenario story description and surely to enhance search accuracy of every picture.

並列關鍵字

Scenario ； Ontology ； Image Content Retrieval ； Scenario Analysis

參考文獻

林穎謙, & 唐玄輝 (2009). 情境故事法對於跨領域合作設計課程之影響. Paper presented at the 明志科技大學技術與教學研討會.

戚玉樑, & 陳彥均 (2009). 發展具有解析使用者認知能力的專家系統於提升資訊查詢效能. 資訊管理學報, 1-24.

蔡宗坤 (2008). 數位學習環境下發展自動化的個人學習路徑規劃系統. 中原大學.

Abowd, G. D., & Mynatt, E. D. (2000). Charting past, present, and future research in ubiquitous computing. ACM Trans. Comput.-Hum. Interact., 7(1), 29-58.

Bardram, J. (2000). Scenario-based design of cooperative systems. Group Decision and Negotiation, 9(3), 237-250.