Feature-point-based multiscale shape coding

In this paper, we propose a new fast and efficient method of shape coding called feature point-based adaptive arithmetic encoding algorithm (FPAE). For dealing with video image, we regard a shape as a set of points that are parameterized by arc length and B-spline bases. Then, the evolution of curve at different resolution levels s in B-spline scale space is achieved by convoluting the curve with the dilated B-spline kernel instead of Gaussian kernel. Compared with Gaussian method, this method has an advantage of fast algorithm. By calculating the curvature, the feature-points including significant information of shape contour can be found. But for shape coding, this shape representation will not very efficient when the shape consist of arcs. The modified shape representation also includes the feature-point that lies in arc and the distance from the feature-point to chord is largest. All of the feature-points are encoded by adaptive arithmetic encoding. Experimental results show that our method reduces coded bits by about 25% compared with the context-based arithmetic encoding (CAE) of the MPEG-4 VM and the subjective quality of the reconstructed shape is better than that of CAE at same Dn.