MPEG視訊分析 - 鏡頭變換的偵測與分類

Abstract

鏡頭變換的偵測是進行視訊分析及檢索的一個重要步驟，其目的是將影片切割成有意義的片段，以利於後續的處理。影片中的鏡頭變換通常可分為瞬間式轉變和漸進式轉變兩種。本篇論文提出一個演算法來偵測影片中的鏡頭變換。針對瞬間式轉變，首先我們利用MPEG資料中Macroblock的資訊做初步的偵測，找出可能發生瞬間式鏡頭變換的候選畫面，同時記錄可能為漸進式轉變的候選畫面群，然後再比對DC值影像的差異來找出真正的換鏡畫面。至於漸進式轉變的部分，我們利用先前偵測瞬間式轉變時所記錄的候選畫面群，來分析DC值變異數的一次微分序列，以提高偵測的效率。

Video shot change detection is fundamental to video analysis and indexing. It is to partition a video into groups of meaningful segments or shots, which are basic units for video processing. There are two basic types of shot transitions: abrupt transition and gradual transition. In this thesis, we develop two-phase algorithm to detect shot transition in MPEG compressed video. For abrupt transition, we use macroblock information extracted from MPEG video bitstream to perform the initial analysis, including the identification of image frames of possible gradual transition. The DC value is then used to find the correct shots. As for gradual transition, we analyze the first derivative of the variance of DC image sequence. Image frames of possible gradual transition obtained in the initial analysis can be used to guide the analysis of the DC image sequence to improve the efficiency.