Video Retrieval by Reranking and Relevance Feedback with Tag-Based Similarity
Takamasa Fujii, Soh Yoshida, Mitsuji Muneyasu · 2018
In this paper, we propose a video retrieval method employing reranking and relevance feedback with tag-based similarity. In the proposed method, we apply relevance feedback to find more relevant videos. Furthermore, we focus on the fact that multiple tags are used to represent video content. Specifically, by vectorizing multiple tags associated with videos based on the word2vec algorithm, we calculate a new tag feature vector. We then formulate the reranking process with relevance feedback as an optimization problem whose regularization terms measure the relevance from both the video and tag features. The integration of the two schemes results in mutual benefits and improves the performance of video retrieval.