應用MPEG視訊分析於電視新聞分段之研究

Abstract

本論文提出一應用MPEG視訊資料中的動態向量及DCT係數之分析來偵測主播鏡頭，而將整個電視新聞節目分割為各段新聞報導之方法，並以電視新聞視訊測試其準確率與處理時間。和其他系統比較起來，本研究的實驗結果達到了相近的準確率；另一方面由於所分析的特性直接取自MPEG視訊，避免了解壓過程中的繁複運算，以及解壓所產生之大量處理資料，主播偵測所用的時間和運算資源遂得以大幅減少。

A method dividing whole TV news program into stories base on anchor person scenes detected by analysis of the motion vectors and DCT coefficients in MPEG-1 or MPEG-2 video is proposed in this thesis. Evaluations of the accuracy rate and processing time are also made with several TV news MPEG videos. Compare to other systems, our method has approximate accuracy rate and achieves an decrease in processing time and resource consuming from avoiding not only a lot of computation needed in MPEG video decompression but also the massive processing data in the decompressed format.