Efficient Video Similarity Measurement and Search

摘要: We consider the use of meta-data and/or video-domain methods to detect similar videos on Web. Meta-data is extracted from textual and hyperlink information associated with each video clip. In domain, we apply an efficient similarity detection algorithm called signature. The idea form a signature for clip by selecting small number its frames that are most set random seed images. then statistical pruning allow fast very large databases. Using ground-truth set, achieve 90% recall 95% precision using only 8% total operations required without pruning. For database around 46,000 clips crawled Web, technique significantly outperforms in recall. show even better performance can be achieved combining them together. Based our measurements, has, average, 1.53 copies.