On the estimation of global motion using phase correlation for broadcast applications

Global motion estimation attempts to find and compensate for camera pan, zoom, rotation or other motions. The majority of motion existing in a video scene consists of local motion coupled with the camera motion; compensating for global motion improves the accuracy of local motion estimation allowing it to search for object motions only. Applications include video compression, background mosaicing and camera vibration reduction. The aim of this paper is to compare different ways of using the phase correlation approach for the estimation of global motion. To investigate this further, we have decided to experiment with three previously undocumented techniques and one that is a variant of the method described by Reddy and Chatterji (see IEEE Trans. on Image Processing, vol.5, no.8, p.1266-71, 1996): (1) phase correlation for rotation and scale (PCRS) on a global basis; (2) PCRS using subsampled images; (3) extended PCRS across multiresolution images; and (4) block PCRS from local estimates