Robust digital image stabilization based on spatial-location-invariant criterion

In this paper, a robust digital image stabilization (DIS) algorithm based on a spatial-location-invariant criterion is proposed to estimate the local motion vectors in jitter videos. In the algorithm, the feature points are detected by sub-region Harris detector and matched by cross diamond hexagon search algorithm, and then in order to obtain accurate local motion vectors, the presented criterion called spatial-location-invariant is used to remove those points that error-matching or in moving objects. With these vectors, the global motion vector is computed by least-squares algorithm in similarity motion model. Finally the real motion model is applied to motion compensation. The experimental results show that the proposed DIS technique can deal with arbitrary rotation, translation, and is robust to foreground moving objects.