鏡頭切換偵測與其應用

A Study on Video Shot Boundary Detection and Its Applications

蔡文宗(Wen-Tsung Tsai)

指導教授 ： 顏淑惠

淡江大學/工學院/資訊工程學系碩士班/碩士(2007年)

https://doi.org/10.6846/TKU.2007.00138

摘要

本論文主要由兩部分組成，第一部份中提出一個有效的方法來正確找出在數位影片中每一個鏡頭(shot)的分界點，而第二部分則是使用我們所提出的鏡頭分界偵測演算法對足球影片進行分割，並對針它們進一步地去做分類。
在數位影片的處理中，無論是作搜尋(retrieval)、分類(classify)或是分群(cluster)，我們都必須以鏡頭為基本單位去作處理，所以一個正確率高的鏡頭分界偵測演算法便變得十分重要，因為它將會影響所有後續處理的效能與正確性。鏡頭的分界主要分成快換景(abrupt transition)與慢換景(gradual transition)兩種。我們首先會根據影片畫格(frame)個別作統計以得到直方圖(histogram)，再計算相鄰畫格之間的相異度，當成之後判斷鏡頭分界的特徵，以此特徵分別對快換景與慢換景使用動態的門檻值去作不同的判斷，之後再找出最足以代表這個鏡頭的關鍵畫格(key frame)。
偵測出鏡頭分界之後，再根據足球影片中所分割出來的鏡頭對去作分類。我們將足球影片分為遠距離鏡頭、中距離鏡頭以及球員特寫這三類，讓使用者能更快速地對影片進行整理。根據所我們提出的方法，對於一般數位影片的分界偵測以及足球影片的分類上皆能夠獲得很高的正確率。

關鍵字

鏡頭 ； 直方圖 ； 快換景 ； 慢換景 ； 畫格 ； 關鍵畫格 ； 影片分類

並列摘要

An accurate and reliable shot boundary detection algorithm is essential to many video processing. In this paper, we propose a new shot boundary detection algorithm and a shot classification scheme on detected soccer video shots.
The proposed shot boundary detection algorithm is based on frame-distance histogram. A rough way to classify the tempo of a video sequence, as static, dynamic, or averaging, is proposed. With estimated content tempo, the notorious problem of the gradual transition cuts detection is solved. Our algorithm is tested on various types of videos and proved to be accurate and successful.
For a soccer video, after shot segmentation has been accomplished, the algorithm proceeds to shot classification into three categories: long shots, medium shots and close-up or other shots. The classification algorithm is based on the grass ratio of the frame. The grass ratio is calculated with motion and color features that no pre-training or updating is necessary. The proposed scheme is tested on various soccer video shots with excellent results.

並列關鍵字

Shot ； Histogram ； Abrupt transition ； Gradual transition ； Frame ； Key frame ； Shot classification

參考文獻

[2]Liang-Hua Chen, Chih-Wen Su, Hong -Yuan Mark Liao, and Chun-Chieh Shih, “On the preview of digital movies”, J. Vis. Commun. Image R. 14 (2003) 358–368.

[4]Hong Lu, Yap-Peng Tan, “Unsupervised clustering of dominant scenes in sports video”, Pattern Recognition Letters 24 (2003) 2651–2662.

[5]Liang-Hua Chen, Yu-Chun Lai, Chih-Wen Su, Hong-Yuan Mark Liao, “Extraction of video object with complex motion”, Pattern Recognition Letters 25 (2004) 1285–1291.

[6]Haoran Yi, Deepu Rajan, Liang-Tien Chia, “A new motion histogram to index motion content in video segments”, Pattern Recognition Letters 26 (2005) 1221–1231.

[7]Young-tae, Kim, Tat-Seng, Chua, “Retrieval of News Video using Video Sequence Matching”, Proceedings of the 11th International Multimedia Modelling Conference (MMM’05).