Visual Image Retrieval by Multiscale Description of User Sketches

This paper describes an application allowing content-based retrieval. The application may be called `sketch-based database retrieval? since the user interacts with the database by means of sketches. The user draws his request with a pencil: the request image is then a binary image that comprises a contour on a uniform bottom. The image-retrieval based on user-provided information such as hand-drawn sketches is a challenging problem in multimedia applications. A human's perception of shape can be rather subjective and has a large variance. In this paper, we propose to represent a sketch and a contours in image database by curvature scale space (CSS) description, which define an invariant distance in the shapes space. In theoretical aspect, the invariance, the completeness and the stability properties of CSS description imply the definition of a metric on the shape space. This distance allows us to compare a sketch to images in database. A prototype system has been developed for image retrieval by sketch. This system takes into account spatial relationships between objects in multi-objects queries.