Research Of Electronic Image Stabilization Algorithm Based On Feature Matching

With the rapid development of computer and Internet technology, image and video as the information carrier are used more and more widely. The instability of video sequence due to the irregular movement of imaging device in acquisition process, not only greatly reduces the video quality and visual effects, but also seriously affects the further image information processing. Therefore, electronic image stabilization as low-cost and high-precision image stabilization technology has become one of the hottest researches in image processing.Principle of jitter in video sequences and the principle and key technology of electronic image stabilization has been studied in this thesis first. It is found that feature matching based algorithm always cost long time due to large computation. Based on the analysis of traditional electronic image stabilization algorithm, this thesis presents an improved image stabilization algorithm to reduce the computational and computation time. Electronic image stabilization algorithm is divided into motion estimation, motion filtering and motion compensation. In this thesis, the key algorithms of these three parts are deeply studied. Specific contents and main results are as follows.In the stage of motion estimation, feature points from images are extracted by the FAST detector with binary search method for automatically determining the threshold. Then they are described by BRIEF algorithm and matched by Hamming distance. In order to deal with the considerable false data in match point set, the symmetry k nearest neighbor matching algorithm with symmetry step is introduced to eliminate most false matches. RANSAC algorithm is then used to filter out local motion matches to calculate a more accurate global motion vector.In the stage of motion filtering, global motion vector is processed by Kalman filter. The scan component is smoothed and the jitter component as the noise is filtered by the way of recursively estimating best system state. This method achieves effective separation of the scan component and the jitter component in global motion vector.In the stage of motion compensation, the backward bilinear interpolation algorithm is used for pixel reconstruction, and the undefined area is filled through combination of the current frame and the reference frame image information. Stable image is no longer missing information.Finally, the proposed algorithm is implemented with VC++, the simulation and verification of algorithm is completed as well. Experimental results show that the electronic image stabilization algorithm presented in this thesis is good enough to meet the needs of image stabilization. Compared with the classical feature-matching-based algorithm, it has better stable effect, less computation and shorter computing time. The improved electronic image stabilization algorithm based on image feature matching achieves an effective treatment on unstable video sequence.This thesis has certain reference significance on feature-matching-based electronic image stabilization algorithm.