Coding of Higher Order Motion Parameters for Video Compression

Due to the ever increasing amount, size and quality of video data, video coding has become indispensable and the demands it has to meet become higher. One important step towards video compression is exploiting the redundancies between temporally succeeding video pictures, known as inter-prediction. As video contents shift with time (i.e. motion happens), motion compensation is required. In current video coding standards, such as HEVC, this is usually realized by estimation and transmission of translational motion vectors, each describing the position shift of a rectangular block. However, when the motion in a video goes beyond vertical and horizontal shifting within the image plane, the translational motion model can only provide an approximate representation. Such cases, for example zoom and rotation, will have to be split into small translational units. The caused block partitioning and imprecise compensation result in an increased data rate and prediction error.

This book is about realizing a concept of motion compensation with higher order motion models. By using additional motion model parameters, more degrees of freedom are available to precisely represent motion forms like rotation and scaling. The main focus lies on the efficient prediction and coding of higher order motion parameters. To this end, several algorithms are presented, each contributing rate reduction to the compression of higher order motion content. When applied to test sequences containing higher order motion, these algorithms achieve an average bit rate reduction of circa 20% over a coder with only translational motion compensation. Hence, higher order motion compensation might be worth considering also for upcoming video coding standards.