新聞視訊搜尋系統

Abstract

隨著資訊的爆炸與多媒體的蓬勃發展，大量的多媒體資訊充斥在人們的生活之中，其中一項重要而且大量的資訊就是視訊，為了使人們在這資訊洪流中能夠更有效的掌握視訊資訊，對視訊內容進行搜尋與檢索，於是乎成為了一項重要的需求。有鑑於此，我們對台灣的電視新聞建構了一個瀏覽網站News Video Browsing System( NVBS)在本論文中，針對新聞視訊內容的搜尋作一探討，建構了”新聞視訊搜尋系統”，我們提出了一項文字分割的演算法，將視訊中的字幕與背景分離開來，再以光學文字辨識(OCR)的技術將其轉換成文字資訊，然後與網際網路上取得之新聞文字稿比對，找出新聞視訊所對應之新聞文字稿，利用此文字資訊作為視訊之索引，以利使用者進行新聞事件，地點，人名的搜尋。本系統利用上述技術對視訊進行自動文字索引，可節省編列文字索引之人力，同時提供使用者在電視新聞內容檢索的功能。

With the dramatic increasing of amount of information and quick development of multimedia technologies, plenty of audio-video data, especially video, have become widespreadly available in our everyday-life. In order to manage these information efficiently, people are in great need of effective methods for video content searching and querying. In this paper, we construct a querying system for video content in TV news and propose an algorithm to extract text from the news video. This method separates caption in the image from the background and translates them into text with OCR technology. After translation, system match extracted text with news scripts retrieved from Internet and use these extra text as index of video contents. With the automatically work of our system, human labors in indexing are totally unnecessarily, and users can search interesting news with text as input, instead of inconvenient, unpractical ways used in immature image-based content searching technologies nowadays.