A robust real-time electronic image stabilization algorithm was proposed based on feature matching and checking. In the algorithm, feature points in reference and current frame were extracted by Kanade-Lucas-Tomasi corner detector and matched by the sum of absolute difference criterion firstly. In the process of checking, a spatial location invariant criterion was presented to delete those points which were error-matching or appeared in foreground moving objects effectively. Finally, the global motion vector was computed by least-squares algorithm in similarity motion model and applied to compensate. Experimental results show that this algorithm is able to meet the requirement in real time and has good performance to the translation, rotation and zoom, and is robust to foreground moving objects.