基于模型的编码

MODEL-BASED CODING

深入研究了基于模型的编码方法，提出了基于内容的网格模型生成方法，通过从边到点的模型生成思想，能得到更好地反映图像结构的模型．在六边形匹配算法的基础上，提出了快速的模型运动估计算法，改进后的运动估计算法在基本上保证了原有运动估计的信噪比的同时，大大地缩短了运动估计的计算时间．模型补偿后的差值图像由一个自适应位分配策略的编码方法编码．实验结果表明，基于模型的编码方法在低码率下能得到较好的解码质量，可用于视频电话等低码率实时编码应用中．

In this paper, an integrity scheme of model based coding is proposed, which includs content based mesh generation, mesh motion estimation and adaptive bit allocation for residual image coding. At first, the idea of Side to Node is utilized to generate the content based mesh model. In order to get mesh model, image is divided into homogeneous texture regions with watershed algorithm and modal nodes are determined with polygon approximation of watershed lines. This model can better represent image structure by the sides of mesh aligning the image contours and edges. Secondly, some improvements of mesh motion estimation based on hexagonal matching are given, such as four step search taking the place of full search, fast texture mapping and novel nodal ordering. The improved algorithm has cheap computational cost in the case of assuring the accuracy of motion estimation. At last, an adaptive bit allocating strategy is proposed to encode the residual image compensated with mesh model. This scheme is used in encoding MPEG 4 standard sequences and the experiment result shows that higher visual quality can be obtained at very low bitrates.