Rate-distortion snake: a tool for optimal shape coding

This paper addresses the problem of joint rate-distortion optimization of spline curves for optimal video object shape coding. It proposes a B-spline snake model which minimizes joint rate-distortion energy to find optimal spline coefficients. A continuous model of contour distortion is employed to model the shape reconstruction error. A coding rate is derived as a continuous function of spline control points. The rate calculation is based on the relative intraframe addressing of control points and includes linear prediction, quantization and arithmetic encoding of the prediction error. The two models enable continuous gradient optimization of snake control points. A global snake energy minimization algorithm is defined which includes snake coefficient optimization as well as quantization step optimization and knot insertion strategy. This results in finding an optimal number and positions of spline control points and optimal quantizer parameters for the given rate constraint.