Research Of Image Stabilization Algorithm Based On Feature Extraction

Image stabilization technology is a widely used video enhancement technology, which is applied in the presence of random jitter of the shooting environment, and it can solve the deterioration of videos. The development of image stabilization technology has experienced three stages: mechanical image stabilization, optical image stabilization and electronic image stabilization. Image stabilization is realized by using computer and image processing technology. Compared with the traditional mechanical or the optical image stabilization system, the electronic image stabilization system has the advantages of easy operation, high stability, high precision, small volume, light weight, low energy consumption and high intelligent real-time processing. In addition, the dedicated digital signal processors’ performance on the market for video or image processing is becoming more and more superior, and they have provided a strong support for real-time video or image processing.The electronic image stabilization algorithm is studied in this dissertation, which involves the principle of the electronic image stabilization, motion estimation algorithm, and common feature detection and description method. In this dissertation, the motion estimation algorithm, including gray projection, optical flow, phase correlation and image feature based method, is presented in detail. The image features introduced including SIFT(Scale Invariant Feature Transform),SURF(Speeded-Up Robust Features),ORB(Oriented FAST and Rotated Brief),BRISK(Binary Robust Invariant Scalable Keypoints),etc, and the application of the open source computer vision library OpenCV(Open source Computer Vision) and Microsoft foundation class library MFC(Microsoft Foundation Classes) to design the graphical user interface software, whose function is two parts: one part of the function is the video framing and changing the size of the video frame, the other part is to achieve a variety of feature detection, description method comparison.On the basis of the above contents, the dissertation proposes an electronic image stabilization algorithm based on ORB features, and it analyzes the performance of the algorithm. When the performance of the algorithm is analyzed, the concept of high quality matching is put forward which is determined by the distance threshold between the feature points. The keypoint is the selection of the distance threshold. In the conclusion of the dissertation the algorithm speed, matching feature points, quality matching feature points and quality matching rate is used in the analysis of the algorithm’s performance. For the selection of the distance threshold, the dissertation gives 4 values to choose from: mean, median, the half of maximum and minimum and 2 times of minimum. The number of high quality matching points obtained by these 4 values is counted, and the experiment show that when the selection is median the result is the best. The conclusion is that the proposed algorithm has the advantage of computing speed, and it can achieve the purpose of image stabilization when there is a translation, rotation and scaling between video frames.