A Study on Global Motion Estimation using Histogram and Weight Map

Abstract

This thesis present a compressed-domain global motion estimation (GME) algorithm that detects global motion change without using an iterative process. Conventional GME algorithms based on the gradient descent and least square approaches are usually accurate and powerful, but are inaccurate in various rate of outliers. Also, these algorithms use an iterative approach, which has a high computational load. To robust the rate of outliers and speed up the GME, this thesis proposes a non-iterative algorithm that exploits an outlier rejection mask and wight map. In simulations the proposed algorithm had highest estimation accuracy and had fastest processing time.