Global motion estimation under translation-zoom ambiguity

Global motion estimation (GME) is an important technique in video processing, segmentation, and compression. In this correspondence, we propose a GME algorithm that offers improved accuracy especially in the presence of large-scale zoom. The proposed algorithm is derived from the observation that there exists an ambiguity between translation and zoom for simple geometric shapes such as lines, which may cause conventional GME to produce inaccurate estimates. Experimental results confirm the improved accuracy of the proposed method compared to the state-of-the-art methods.