As the rapid popularization of digital imaging equipment, video character recognition becomes more and more important. Compared with traditional scanned document, characters in video document usually suffer from great degradation and meet trouble in recognition. Thus, a systematically study of video degradation will be very useful for video OCR. In this paper, a video degradation model is proposed to imitate the process of video character image generation. The generated character images are used to make synthetic dictionaries to improve the recognition performance of real degraded characters in e-Learning videos. Experiments on 24317 e-Learning video characters prove the effectiveness of our method.