Novel video stabilization for real-time optical character recognition applications

This paper presents a novel video stabilization algorithm for real-time optical character recognition (OCR) applications. The proposed method generates output frames in order to stabilize the position of a target word that will be recognized by the OCR application. Unlike in conventional algorithms, in the proposed algorithm, a causal low pass filter is not applied to the trajectory of the target word for reducing the high frequency component of camera motion. The proposed algorithm directly calculates the stable position of the word using two forces: the force used to pull the target word to the center of an output frame and a back force used to return the center of an output frame to the center of an input frame. Hence, the proposed algorithm significantly minimizes the time take to respond to sudden camera movement. Although the proposed method may not outperform state-of-the-art video stabilization in terms of video stability, the proposed technique is much more appropriate for real-time OCR applications than the conventional techniques in terms of accuracy, computational cost, processing delay, and the time taken to respond to sudden camera movement. Simulation results prove the superiority of the proposed method over conventional techniques for real-time OCR applications.