This thesis concerns two popular motion estimation methods for video coding: a gradient-based and a matching-based method.
While gradient-based methods have been widely studied, little attention has been paid to their convergence behaviors especially from the standpoint of coding applications. In the first part of the thesis, we quantitatively analyze those characteristics mainly on convergence speed and steady-state estimation error, in case that gradient-based methods are used for estimating 2-dimensional motion from the region of the image where translational motion occurs. All analyses are based on an isotropic exponential covariance image model, and estimates are derived with respect to the real displacement and the second order statistics of the region. By the obtained results, we show the properties of two different spatial gradient measures and the effects of noise corruption and prefiltering. In addition, we analyze the situation that two objects with different motion are inside the region from which we would estimate only one motion information. For that case, convergent direction and steady-state estimation error are considered.
In the second part, we propose an efficient motion compensation technique based on block-matching algorithm(BMA) and global motion information. Due to the limitation of BMA which can describe only translational motion, when the more complex motion such as rotation and zooming occurs, its prediction efficiency is degraded. Proposed algorithm extracts global motion parameters from a motion vector field(MVF) generated by BMA, and obtains a new MVF by the use of the globally motion-compensated image. And then, a motion vector from either MVFs is adaptively transmitted for every block. By this concept, we can expect the improvement of prediction gain and the reduction of motion information owing to the parametric extraction of global motion. Simulation results show that the proposed algorithm is better than BMA in terms of both prediction gain and motion information for test sequences.
