Dynamic Motion Vector Searching Algorithm Using Window Adjustment, Global Motion Information, and Direction Priors

Motion vector (MV) searching plays an important role in video compression and object tracking. In this work, an advanced MV searching algorithm is proposed. The key idea is to adjust the searching window according to the global motion vector of previous coded frame. The movement of a video can be classified into object motion and camera movement. That is, total motion = object motion + camera movement. The camera movement introduce the global motion vector, which is usually the same for any part of a video. Moreover, for most of videos, the direction of object motion usually distributes non-uniformly. For example, the video taken from the roads has much more horizontal motions than vertical motions. Based on these facts, in this work, we proposed an effective MV searching algorithm using global motion information and direction priors to design a dynamic searching window. Different from existing works, which using the same searching range in all directions, in this work, we make the search range configurable in four directions. Simulations show that two-thirds computation is saving by proposed MV searching scheme with only 0.20db PSNR quality degradation in average.