Aplicación de la realimentación por relevancia con base textual para sistemas de recuperación visual

Pseudo-relevance feedback is a common technique in information retrieval. Content Based Information Retrieval systems (CBIR), like visual retrieval engines, suffer from low performance due to the distance between the semantic interpretation of an image and its visual features. This is the reason why such a systems are empowered when textual information is associated to the images in the collection and the CBIR is combined with a textual retrieval system. In this paper, some experiments on applying text-based pseudo-relevance feedback on CBIR are reported. The results obtained show that these techniques may enhance the behavior of traditional fusion approaches.