Video object extraction with interactive image cutout and active search

This paper presents a video object extraction method. To cutout moving objects from a video, we adopt an approach of combining an interactive image cutout and an active search. In the proposed method, a user draws seed lines on the object and the background respectively in the first frame of a video. Using these seed lines, the image cutout algorithm automatically separates the object from the background. To the following frames of the video, the active search algorithm estimates the changes of position, scale and orientation of the object region. The seed lines are affine-transformed according to the estimation. To the new seed lines, the image cutout is performed again. Repeating the image cutout and the active search, dynamic object region can be accurately separated from each frame. Experimental results have shown that the proposed method is robust to affine transform of object's shape. Besides, the results have shown that the method can deal with non-rigid object and partial occlusion.