Study Of The Algorithms For Electronic Image Stabilization Systems

Electronic Image Stabilization (EIS) is a new methed to stabilize image sequences with computer image processing technique.It has a lot of advantanges,such as high image stabilizing precision,compact size,lightweight,low power consumption and reasonable price. Electronic Image Stabilization (EIS) technique is a growing research focus, in which motion estimation and motion compensation are of the core techniques. The research on increasing efficiency and accuracy of motion estimation and distinguishing camera scan from dithering to realize a real time and accurate compensation becomes a hot topic at present. However, it leaves us a problem to establish a standard criterion to give an objective evaluation on the stabilization result.In this thesis, the author emphatically discusses the motion estimation and compensation algorithms, and makes a summary of their application areas. Firstly, to deal with camera translation, an analysis of Block-Matching Algorithm (BMA), bit-plane matching algorithm (BPM) and Projection Algorithm (PA) is made and then the Projection with Motion Correction Algorithm (PMCA) is proposed. Motion estimation is obtained only by making a correlation operation on the projections of columns and lines. Moreover, motion correction is applied to prevent error at the beginning of motion compensation propagating to the successive frames. Experimental results show that the proposed algorithm can effectively smooth out the unwanted camera translation motion and follow the intentional camera movement. Secondly, to deal with camera rotation and zooming, an algorithm based on feature matching is proposed: Features Matching with Validation Algorithm (FMVA). The edge points are selected as features and then each feature window with the specified point centered in it is matched. In particular, the matched pairs of points are validated based on the distance invariant criterion to maintain high computing accuracy. The difference of the current motion parameter and the smoothed motion parameter is the compensated motion vector and then compensation is applied on each current frame to make it stabilized. Experimental results show that FMVA can effectively smooth out the unwanted camera rotation motion and make a real time stabilized video sequence.