To overcome the undesirable shakes of a camera and to implement the image stabilization in a real time, a digital image stabilization approach based on local feature points was proposed. Local feature points have been widely investigated in solving problems in image processing, pattern recognition and computer vision, such as feature matching, object detection, target tracking and image navigation. First, the Harris algorithm is proposed to select key points in the image sequence where image motion happened due to vehicle or platform vibration, and Normalized Cross Correlation (NCC) matching algorithm is used to match local feature points between reference frame and current frame, here, a novel bi-directional search strategy is proposed to extract the more robust feature points. Second, the Random Samples Consensus (RANSAC) algorithm is adapted to eliminate the wrong matching points, and the reserved matching feature point is taken into affine motion model to yield global motion vector. Finally, jitter parameters are extracted from global motion vector, and original image sequence is compensate. The experimental results demonstrate that the proposed DIS approach can process affine motion of image sequence fast and efficiently, accuracy lower than 1 pixel, and have a better robustness. This approach can be used in moving vehicle cameras system.