A new motion-compensated technique for video compression

In this scheme, image triangular partitioning is adopted instead of the usual rectangular partitioning. Thus, various types of motion, such as camera zooms and pans, object rotation, and local deformation, can be well modeled by the affine movements of the triangle patches. Therefore, motion estimation is to compute the displacements of the vertices of triangles. Two approaches for handling this problem are given. Finally, a motion-compensated scheme based on the texture-mapping technique is introduced. Compared with the classical block matching algorithms, the proposed scheme has the following advantages: (1) it can compensate various types of motion besides the translation motion; (2) it can offer a motion vector for every pixel rather than a motion vector for the whole triangle; and (3) it provides a possibility of reducing the annoying ´block effect´.<>