Interactive model-based matching retrieval

Abstract

We present a system for image retrieval based on the query-by-sketch technique. In our system the user formulates a query by drawing a sketch of the shape of the object (the "model") he is looking for and requests all images, stored in a database, containing such similar object. The system evaluates the similarity between the shapes in the database and the rough sketch by applying, sequentially, three similarity criteria based on the shape description. The system tries to emulate some typical human features like the power of visual communication, the fast cognitive feedback loop and the relatively small memory load for a natural and inexpensive human-computer interaction.