Tag suggestion and localization for web videos by bipartite graph matching
Wei-Ta Chu, Cheng-Jung Li, Yeh‐Kai Chou · 2011
In this paper, we formulate video tagging as a bipartite graph matching problem. Starting from existing tags that were originally provided by video owners, we conduct keyword-based image search on Flickr. Tags associated with the retrieved images are collected as candidate tags for tag suggestion. Relationships between keyframes extracted from the same video shot and candidate tags are then described as a bipartite graph, and best matching between two disjoint sets is accordingly determined to suggest new tags to this video shot. In constructing the bipartite graph, visual characteristics in terms of the bag of word model and tagging behaviors are jointly considered. Experimental results demonstrate that the proposed features and methodology achieves superior performance over previous approaches.