Research On Video Image Processing Based On Electronic Image Stabilization Technology

With the rapid development of computer science and the widely application of mobile camera, image and video become important transmission medium of information. However, in the actual work enviroment, the camera carrier is often difficult to avoid unpredictable random shaking, This shaking can cause the blur of the video, resulting in human visual fatigue. Due to this, we adopt electronic image stabilization technology to eliminate the noise jamming. With its advantages of low-cost,high-precision of electronic image stabilization, it has important theoretical significance and practical value.SURF(Speeded Up Robust Features) is a fast and robust image registration algorithm based on SIFT(Scale Invariant Feature Transform). Due to the invariance of scale and rotation of feature points, it is commonly used in image matching. However, on the one hand, in practical applications the isolated point and the noise point may cause mismatching points, on the other hand, SURF and SIFT feature points record the ralationship of different scale between the feature point and around it, so easily caused the described similar bitween the different image feature point and it can be matched each other after extracting their SURF feture points. In order to solve the problem, this thesis presented an electronic image stabilization method of combing K-means clustering method with SURF. At first, SURF is fast computed based on the integral image and through the Hessian detector, the feature points are extracted. Then for each point the dominant orientation is assigned by computing Haar wavelet responses,and then the descriptors are generated. The feature point matching was implemented with nearest matching approach which could reduce matching points for accelerating the matching speed. Then K-means clustering method was used to classify the matching point and eliminate the abnormal point, achieving the secondly accurate feature point matching. The experiments show that the algorithm can effectively reduce the influence of isolated point,the noise point and structure irrelevance, can improve its accuracy.The image effects and quality were more superior compared with troditional methods. Moreover,combination of two algorithm was an new trendency in image processing.