Research On Digital Image Stabilization For Handheld Imaging Devices Under Compound Motion

Digital image stabilization(DIS) enables to eliminate annoying unwanted motion fluctuations from videos captured by unstable optical systems, using image process techniques. Basically, DIS contains three phrases: motion estimation, motion filtering and motion compensation. In this paper, by analysis of the above procedures, we aim to propose a fast image stabilization solution for handheld devices.We first establish the image motion model caused by camera motion, and then briefly introduce the basis and challenges in DIS. Afterward a novel equivalent motion model is proposed to fit the compound inter-frame motions which combine arbitrary rotation and large-scale transition in the process of handheld camera imaging. Besides, the advantage of this model in the implementation of DIS process are expoundedIn DIS motion estimation, we focus on the feature points tracking method and gray feature matching method, respectively. Based on the proposed equivalent motion model, a novel image motion estimation approach is proposed. It uses the image pyramid to reduce the computing burden, as well as a modified hexagon-search strategy to improve the search capability. Furthermore, an eliminating strategy is performed to remove the incorrect local motion vectors obtained in estimation process.In motion filtering part, we analysis three fitting or filtering techniques, namely Bezier fitting, mean filtering and Kalman filtering approach. The implementations of the above methods in the real time DIS are discussed, and results indicate the evident advantages of Kalman filtering in terms of both filtering performance and real time. Additionally, the motion compensation part mainly discusses two types of compensation patterns, as well as the image interpolation during the compensation process.Finally, a fast image stabilization algorithm based on the equivalent motion model is introduced, which consists of the proposed motion estimation method, Kalman filtering and pixel-accumulation based compensation. Implementation under MATLAB environment indicates that our proposed method outperforms some currents algorithms in terms of valid motion stabilization scope, time complexity, thus meets requirements of DIS system for handheld devices.