A Novel Global Motion Estimation and Compensation Framework in Compressed Domain for Sign Language Videos
Anjan Kumar Talukdar, Manas Kamal Bhuyan · 2020
In this paper, we have proposed a novel framework for Global Motion Estimation (GME) and Global Motion Compensation (GMC) in compressed videos. The framework employs the motion vectors (MVs) partially decoded from the compressed videos which are inherent in the videos. GME is a challenging task in the presence of moving objects in a video as the estimation of the Global Motion (GM) parameter vector is affected by these moving regions. In this framework, we have proposed a novel method based on the histogram of magnitudes of MVs for removing the moving regions from a video. The MVs corresponding to background Motion Blocks (MBs) are selected for the estimation of Global Motion (GM) parameter vector from the bin corresponding to the peak value of the histogram. After obtaining the GM, finally compensation of the GM is done by removing this GM from the original MV field. We have tested our algorithm for American Sign Language (ASL) videos encoded with H.264/AVC JM encoder and results are found to be satisfactory in presence of various outliers.