A multimodal interactive image information retrieval system for smartphones
Place: Large Lecture Room - CVC
Affiliation: Universitat d'Alacant. Spain
Advanced smartphones can provide specific metadata to reduce the search space in computer vision tasks. For instance, local time can be used to search only those available pictures with similar light conditions (like Goggles does). Novel metadata are proposed for this task, like the angle in which a photograph is taken, which can give information about the object size. A smartphone prototype has been developed to gather information about objects from a photograph by combining metadata with image features. The method is interactive, and the user can validate the response, adding objects to the database and contributing to improve future search results. A joint project is proposed to further develop this application, in which CVC experience with image feature extraction and machine learning methods applied to computer vision tasks could be very helpful.