Image Signature Robust to Caption Superimposition for Video Sequence Identification

摘要: This paper proposes an image signature robust to caption superimposition for video sequence identification. A new which is a set of local features developed high-speed frame-by-frame matching sequences. The frame obtained by partitioning the into blocks and extracting feature representing dominant type edge direction from each block. similarity between signatures calculated comparing types corresponding blocks, counting number having same type. weighting scheme based on probability block can be applied calculation improve performance. experimental results identification show that proposed achieves precision 99.65% recall 99.45%, improving both more than 30% compared with conventional signature.