Review of 2D shape descriptors based on the curvature scale space approach

Shape representation and description is one of the major problems in Content-Based Image Retrieval (CBIR). Many reviews of 2D shape descriptors have been proposed using different taxonomies. Generally those reviews consider that there are two families of shape representations: Region-based and Contour-based techniques. Among the well known contour based techniques, the Scale Space representation (CSS) approach introduced first by Asada and Brady and extended by Mokhtarian et al. as CSS map descriptor. The CSS map has the advantage of being invariant to scale, translation, and rotation, and is shown to be robust and tolerant of noise. In addition, it has been adopted by MPEG-7 standard. However, this latter presents some drawbacks. Several important techniques based on the CSS approach have been developed to present a remedy for the drawbacks of CSS descriptor. In our knowledge, up today there are not reviews which treat especially those techniques based on the CSS approach. In this paper, we propose a short review that classifies and summarizes these important techniques following a new taxonomy dealing especially with all techniques based on the CSS approach. Their advantages and disadvantages are presented, and some recent research results are also included and discussed.