PhotoMole: retrieval from a database of natural images

This demonstration extracts images from a set, based on their similarity to a query image. The scenes handled are of various types: indoor or outdoor, natural (trees) or structured (furniture). The constraints on the shooting are weak, allowing viewpoint changes, partial occlusion and illumination changes. Different sensors can be used to capture the database images and the query image: we tested it on digital cameras, webcams and low-resolution mobile phones. The retrieval is fast (< 1/3 second) and images can be added dynamically to the database.

[1]  Cordelia Schmid,et al.  Scale & Affine Invariant Interest Point Detectors , 2004, International Journal of Computer Vision.