Extreme curvature scale space for efficient shape similarity retrieval

The description of the object shape is an important characteristic of the image; several different shape descriptors are used. This paper presents a novel shape descriptor which is robust with respect to noise, scale and orientation changes of the objects. It is based on the multi scale space approach to identify shapes. The descriptor of a shape is created by tracking the position of extreme curvature points in a shape boundary filtered by low-pass Gaussian filters of variable widths. The result of this process is a several contours map representing the extreme curvature points of the shape as it is smoothed. The maxima of these contours are used to represent a shape. We demonstrate object recognition for three data sets, a classified subset of database SQUID, the set of silhouettes from the MPEG-7 database and the set of 2D views of 3D objects from the Columbia Object Image Library (COIL-100) database. The results prove the performance and robustness of the developed method and its superiority over Curvature Scale Space (CSS) in shape with shallow concavities.