Advanced contour compression algorithm using weighted curvature, Lagrange curve approximation, and improvement adapitive arithmetic coding

Contour compression is to encode the boundaries of objects and important for object-orient image compression and cartoon-like image compression. In this paper, we propose an advanced algorithm to perform lossy compression for contours. First, we find that among all curves, the 3rd order polynomial is most suitable for contour compression. Second, we suggest that, instead of encoding the coefficients of curves directly, it is more efficient to encode the heights of curves. Moreover, we apply an alternative definition of curvature to find the dominant points. Furthermore, we apply the improved version of adaptive arithmetic coding, including increasing the probabilities of the values with the same sign or with similar amplitudes, to encode the heights of curves. Simulations show that, when the error is similar, the proposed algorithm requires much less number of bits than other existing methods for contour compression.