基於關聯規則之標籤分類影音檢索系統

Abstract

近年由於線上分享影音資訊快速的發展，五花八門的影片在網路平台上隨處可見。雖然在網路上觀看影片很方便，但是都只能搜尋一部完整的影片，網路上很多長度都至少超過十分鐘的影片，如果只是想瀏覽影片的特定片段，或是從某個特定時間點開始瀏覽，就還要花時間在尋找正確的時間點上。 本研究主要以Android、JSP、XML、OWL等技術實現標籤分類影音檢索系統。利用OpenCV來分析影片內容，找出關鍵畫面並做為影片的分段點，自動為整部影片分割成多個片段。系統提供使用者輸入影片或片段的相關的資料以及與其內容相關的標籤，並以XML的格式儲存相關資訊，並且系統會將所有影片及片段的標籤利用關聯規則(Association Rule)來計算影片的相互關係，並以OWL格式文件儲存。使用者查詢時利用XQuery查詢語言查詢資料庫中影片及片段內容，以及利用Portege API將OWL文件中儲存的標籤關聯資訊影片列出，讓使用者除了能夠快速地找到想要瀏覽的影片內容，還能夠額外列出和使用者原本想找的影片類別相關的其他影片。分段功能及關聯分類功能的結合，可以提供強大的功能。例如：語言學習的影片，使用者可將不同單字的學習做分段，並且加上主題標籤，如此一來除了可利用標籤關鍵字來搜尋欲瀏覽的片段之外，還可以看到其他使用者上傳的與關鍵字有關聯的影片或片段，學習到更多的相關單字。Due to the rapid growth of sharing audio and video contents online in recent years, various of videos can be seen on the Internet. Although watching videos on the Internet is very convenient, but only in full video searching. On the Internet a lot of videos the duration is tens of minutes, if people only want to watch a particular clip, it will spend a lot of time to search the right way. In this thesis, we use Android、JSP、XML、OWL to develop the A Hashtag-Classified Video Query System. We use OpenCV to analyze video contents, find the key frame and be the time of video segmentation, finally split the whole video into multiple segments automatically. System allow users to input the relevant data or hashtags for video or segments,and store the data use XML format. We use the native XML database eXist to store XML documents, and system will compute the relationship between the videos using Association Rules according to the hashtags form all videos and segments, and store the result into a OWL format file.System use XQuery to query the data of videos and segments when users searching, and use Portege API to get the relevant videos from OWL file. So users not only can find the video or segment they want to watch, but also can watch the other videos relevant to the original key word. Combining the splitting and asscociation rule function, it will be a strong system.For example, user can split a language learning video according to different words, and add hashtags to them.As a result, user not only can use key words to search the segment they want, but also can browse other segment related to the original video uploaded from the other users, improving the efficiency of language learning.