- 學位論文
以顏色為基礎的卡通動畫視訊資料庫內角色搜尋與檢索
Color-based Character Search and Retrieval in Cartoon Video Database
摘要
目前許多視訊分享網站所使用之檢索方式多依照較為傳統的關鍵字搜尋方式，自各視訊的標題或標籤(Tag)進行比對取得使用者所需資訊，對於要查詢某個特定角色出現或不出現，或是回憶某些經典片段的使用者，在查詢結果中仍須作線性搜尋仍有些不便。本研究主要探討一套卡通動畫視訊檢索系統，利用物件偵測與視訊索引的相關技術，偵測卡通角色在每部視訊的像框內出現狀況，並以此記錄該卡通角色在卡通視訊內出現的時間起訖點，進而以此建立視訊片段。系統亦利用影像比對的相關技術，能夠讓使用者直接輸入欲查詢之卡通角色之名稱或圖片後，透過影像比對之相關技術，判別出使用者欲看之卡通角色為何，並進入資料庫中查詢，傳回使用者所需之卡通角色有出現之視訊與較精確之片段。
並列摘要
Query by keywords is the most popular approach used by most video websites such as YouTube, NicoNico, etc. Keywords are used for matching the tags and titles of the video streams set by the owners, but it is not convenient for query for video clips where the character we want to watch was shown. We discussed a video clip query system which is available for searching clips where a certain character appears. This system allows user to use keywords or an image which contains a cartoon character for query certain video clips by comparing color histograms and calculating the similarity. Video clips are built by searching the points when the cartoon characters start to appear and disappear in video streams using object detection methods and video indexing methods, and these points are discovered by observing whether the cartoon character appears in each frame of a video stream or not.