Research On Electronic Image Stabilization Algorithm

Electronic image stabilization technology as a means of stabilization,has now been widely used in the screen processing.Screen shooting,inevitably is affected by the impact of jitter which is the translation,rotation,zoom and so on.By the impact of jitter,the screen will appear fuzzy,ghosting,jumping and other phenomena,causing visual fatigue.The traditional image stabilization system concludes mechanical image stabilization system and optical image stabilization system.Both are accountable and expensive.Electronic image stabilization is mainly through the digital image processing method to achieve the image,which determines the technology lightweight,flexible,low price,small size,easy to achieve through the software features.The electronic image stabilization mainly includes three aspects: motion estimation,motion filtering and motion compensation.The most important is the motion estimation,which determines the precision and real-time of the whole image stabilization system.This paper introduces the spatial model of electronic imaging and the mathematical model of the mutual transformation between objects in two-dimensional plane,which provides the theoretical basis for the following motion compensation.Several typical image stabilization algorithms are discussed,and their advantages and disadvantages are pointed out.In the aspect of motion filtering,the mean filter and Kalman filter are introduced,and the working principle and advantages of Kalman filter are introduced.This paper focuses on the motion estimation,mainly divided into two classes,simple translational motion estimation and complex rotation,scaling and other motion estimation.In the small translation movement,the block matching algorithm is mainly studied.For the most time-consuming search strategy of block matching algorithm,several typical search strategies are studied,and the improvement is put forward on the basis of adaptive search strategy.When not in the center,according to the location of the smallest point to choose to use oblique cross search,or cross search,instead of the original when the minimum point is not in the center of the cycle search.This reduces the number of search points and speeds up the calculation.In the complex motion estimation,the image features are used to realize the image stabilization.This paper focuses on the SIFT algorithm,and introduces the basic steps of SIFT algorithm in detail.Aiming at the problem that the sub-dimension of SIFT algorithm is too large,the feature matching is too slow and the main direction and rotation are calculated in the process of forming the feature descriptor.The eigenvector is used to replace the original vector with a gradient of 128 pixels,and the main direction is calculated by using the anti-rotation of the ring and the direct cyclic eigenvector.Finally,the experimental results show that the two improvements proposed in this paper have a good effect.