Visual image retrieval by elastic deformation of object sketches

We present a method for image retrieval based on sketches of object shapes. In our method, the shape is drawn by the user, and instantiated as an elastic template. The template is then deformed to match as well as possible the objects in the images. The degree of match achieved, and the elastic deformation energy spent to achieve such a match are used as a measure of the similarity between the template and the image object. Images can be ranked according to the matching of their objects with the sketch, and the top ranking ones can be presented to the user for browsing. The elastic matching is integrated with arrangements to provide for scale invariance, to take into account rotations and spatial relationships between objects for multiple-object queries