Accuracy, Rate and Complexity Tradeoffs of Transformation Estimation using Lie Operators

摘要

The accuracy of the conventional motion estimation based on translation motion vectors could be improved by integrating the Lie operators, which are capable of capturing potential object transformations such as scaling, rotations and deformations in video scenes. The increased accuracy of the motion estimation comes at the expense of added computational complexity associated with estimating the additional parameters that describe the transformations such as degrees of scaling and rotation angles. Extra bits have also to be spent on coding these transformation parameters, in addition to the bits required for coding the conventional motion vectors in a typical video coding system. In this paper, we introduce three different methods of transformation parameter estimation. We explore the tradeoffs among the accuracy, bit rates, and the complexity of Lie-operator based motion estimation algorithms by conducting a comparative study on these three methods.