Video Search with Context-Aware Ranker and Relevance Feedback

Interactive video search systems effectively combine text-image embedding approaches and smart user interfaces allowing various means of browsing in intermediate result sets. In this paper, we combine features from VIRET and SOMHunter systems into a novel approach for segment based interactive video retrieval. Based on our SOMHunter log analysis and VIRET tool performance in known-item search tasks, we focus on two specific features - a combination of context-aware ranking by text queries and Bayesian-like relevance feedback approach for refining scores using promising candidates.