在新聞影片中,字幕往往指出了此段新聞的要點。因此字幕在新聞影片裡扮演著一個重要的角色。本研究的目標在於將字幕由新聞影片中偵測並且擷取出來。研究中將使用SVM演算法做為分類器。SVM藉由分析頻率的特徵將字幕候選的區段粗略的分類出來。經過SVM後即可找出字幕候選遮罩,同時利用這些遮罩找出字幕開始之畫面減少處理的時間,因為它可以避免處理到重複的字幕。最後再利用SVM將字幕由每張字幕開始畫面之字幕候選遮罩偵測出字幕之面積。本實驗提出了一個低計算耗費並且有效之新聞字幕定位與擷取系統。由實驗結果可知,字幕將被準確的由影片中尋找出來。
Captions summarize the information of the news story. Thus captions play an important role in news video. In this paper, we aim to localize and detect the caption area in news video. The support vector machines (SVM) was used to classify the caption area and non-caption area coarsely by analyzing the frequency features. After the SVM was performed, a candidate caption mask for every video frames was identified. Then the caption begin frames were detected to reduce the number of processing frames. Finally, the SVM was used to detect the caption area in the candidate caption mask from the caption begin frames. This paper provides an effective method for caption localization and detection in news videos with low computation cost. The experiment results show that the caption could be found accurately.