Virtual Reality for Image Retrieval

Abstract

Image database systems have been widely studied in the past years. Having to deal with visual information, a great effort has been put in designing and implementing user interfaces for the different retrieval engines. Most of the interfaces proposed so far allow the user to specify the query through visual examples, either by feeding a complete image to the system or by drawing some colour patches on a sketch-pad. However, formulating a query from scratch remains a hard task and users tend to use existing images as examples. Virtual reality provides realistic reproduction of natural environments along with a wide set of previously unexplored interaction techniques. We suggest to use non-immersive VR environments as the interface to perform query by content in a natural and effective way. Our non-immersive VR-based approach permits the user to navigate and edit a virtual environment and also to take photographs. These are then used to query a database of images based on their content. Relevance feedback interaction allows the user to extract parts of retrieved images and apply them back to the virtual environment from which a new photograph can be taken.