以時間化之邊緣型態表示法為基礎之影片資料擷取

Abstract

視訊影片資料庫的有效性依靠於是否能夠容易找到有興趣的視訊影片。為了提供能根據視覺的內容來探索、瀏覽和擷取影片的能力，在視訊影片資料庫中有效的偵測相似影片的技術成為在多媒體與資料庫相關領域中的一個最重要與富有挑戰性的議題。我們提出了一個新的時間化之邊緣型態表示法以用來偵測影片的相似度。為了減少計算複雜度，我們使用影片中的關鍵畫面來做比對。利用一個以GOP為基礎的方法來做場景變換偵測，並將影片切割為許多片段。從已切割好的影片片段中選出I-frame作為其關鍵畫面。最後，使用一個結合梯度能量直方圖、最高邊緣密度垂直切片變化與區域逆機率差異矩的混合邊緣特徵表示法來計算影片間的相似度。從實驗結果可以看出所提出的架構相當實用並可達到很高的準確率。

The usefulness of a video database relies on whether the video of interest can be easily located. To allow exploring, browsing, and retrieving videos according to their visual content, efficient technique for detecting the similar video sequences in the video database has become one of the most important and challenging issues in multimedia and database related areas. A novel temporal edge pattern descriptor is proposed for video similarity detection. To reduce the computational complexity, we match the key frames of the query sequence and the sequences to be matched. A GOP-based method is used to detect scene change, and the videos are then segmented into shots. I-frames are selected from the segmented shots as the key frames. Finally, a hybrid distance measure is used for video similarity detection which is combined with the distance of gradient energy histogram, highest edge density vertical slice (HDVS) variance and region-based difference of inverse probabilities (RDIP) moments. Experimental results show that the proposed scheme is quite robust and achieves high performance in precision-recall rate.