Object segmentation with affine motion similarity measure

Affine motion model is widely used in motion segmentation. Accurately estimating and evaluating affine parameters are two key problems in such kind of approaches. This paper tries to address these issues by presenting an effective video object segmentation method based on affine motion similarity measure. The image is firstly segmented into irregularly-shaped intensity homogenous regions. Then, relatively reliable affine parameters for each region are estimated by a robust motion estimator according to the individual coordinate system of each region. Finally, a new motion similarity measure and merge process is applied to obtain meaningful objects. Experimental results demonstrate the effectiveness of the proposed method.