Video Retrieval Based on Periodic Feature Combing Dynamic Time Warping and Earth Mover’s Distance

• Main Authors: Meng-Ju Tsai, 蔡孟儒
• Other Authors: Mau-Tsuen Yang
• Format: Others
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/80428177742243667267

碩士 === 國立東華大學 === 資訊工程學系 === 101 === With the rapid advance in digital imaging, video sharing websites become popular such as YouTube and NICONICO. Compared to images, videos contain a lot more spatial and temporal information. The purpose of video retrieval is to accurately find videos online according to the needs of users. Although numerous video retrieval algorithms have been proposed, periodic information hiding in videos is seldom utilized. We propose a video retrieval method based on the analysis of periodic features in videos. The Earth Mover’s Distance (EMD) is used to measure the difference of periodic features between two videos. The Dynamic Time Warping (DTW) is applied to match two videos allowing stretching and shrinking in the temporal domain. Most query-by-example video retrieval systems try to find videos with the same contents. With the consideration of periodic features, we aim to find videos of an identical event, but captured from distinct viewpoints. Also, we plan to discover videos containing a typical dancing style, but playing in various tempos.