Low-complexity global motion estimation based on content analysis

Abstract

In this paper, we present a new approach for global motion estimation based on pre-analysis of the video content that works with compressed video data (e.g. MPEG-2). The novel idea is to perform pre-analysis of the scene content based on DCT coefficients without explicit MC-image reconstruction. Here, with the use of affine model and the combination of the spatial and temporal continuities of global motion parameters, accurate and robust dominant motion can be estimated and the outliers can be removed. Moreover, the problem of inaccurate initial descending direction estimation associated with the classical global motion estimation methods has been overcome. A global motion model, providing real-time performance, is achieved from decoded P-frame motion vector fields.