(A) study on motion estimation and motion compensation algorithms for the image sequence with global motion

Abstract

Global motion caused by camera zooming, panning, and rotation is quite common in video sequences. Since global MC(Motion Compensation) itself is unable to compensate for local motion caused by the movement of object in front of camera, the local MC is often incorporated with global MC. It has been shown that in video compression, global MC can significantly reduce the residual of MC and entropy of local motion vector field. But the scope of the previously reported works on the global MC does not exceed the empirical confirmation of the effect of global MC on prediction accuracy and motion bitrate. Since the global MC and local MC are always used together for the practical MC of video sequence, the relationship between two MCs and between global and local motions should be investigated to fully exploit the global motion for the video compression. In this dissertation, we have studied the issues related to ME (Motion Estimation), MC, and motion information coding methods for the video compression using global motion. First, we analyze the motion of video sequence with global and local motions and formulate an analytic description of such motion in a closed form. Based on the description, the well-known motion models such as the composite and translational models are compared in terms of motion accuracy and motion entropy. Note that the composite model consists of global motion model and local motion model and its functional form is the sum of the two motion models. For the comparison on the motion accuracy, MAE(Motion Approximation Error) which is the MSE between the true motion vector and modeled motion vector is introduced to evaluate the performance of the composite and translational models in approximating the true motion vector, in other words, prediction accuracy. Also, the motion bitrate necessary for representing these models are compared in terms of motion entropy. From the analysis, we show that there is a certain point which classifies the relative p...