Shape classification based on histogram representation in curvature scale space

Shape classification is one of the central topics in computer vision. Due to the robust shape representation, curvature scale space (CSS) has been adopted as the default in the MPEG-7 standard. One of the drawbacks of the standard CSS algorithm is its complicated and time-consuming search for all possible ways of aligning the contour maxima from both CSS images by shifting and mirroring the Query or Gallery image. In this paper, firstly, we quantify the CSS descriptor by transforming the CSS image to circular vector map. And then define two histograms by dividing angle and radius into equal interval individually in polar coordinate system. Finally, we make use of the city block distant measure to obtain the similarity between two shapes. The advantages of our proposal are its simplicity and execution speed. A classification of shapes evaluation procedure shows the new method's efficiency.