Abstract
For scalable and model-based coding of a videophone sequence at very low bit rates, a 2D scalable face model has to be designed. An efficient algorithm is presented for the automatic detection of a chin contour in a human face. The chin is first represented by a deformable template consisting of two parabolas. Then, a cost function is minimized to find the best fit of the template to the chin. Finally, the chin contour is detected using the active snake algorithm, which is initialised by the best fit of the template. In order to obtain the smoother external force of the active snake model, gradient vector flow (GVF) is used, which is derived from the edge distribution. Experimental results show that the proposed method can drive the snake model to the most optimal chin position. This method can be used for 2D scalable face model design and 3D face model adaptation.