一种多子块协同的自适应快速数字稳像方法

A Fast and Adaptive Digital Image Stabilization Method Based on Multi-block Collaboration

针对现有数字稳像系统稳像精度低、难以实时处理的问题,该文提出一种快速鲁棒的数字视频稳像方法。采用亚像素相位相关对图像子块进行局部匹配,根据局部运动的一致性,利用多子块协同获取全局运动矢量,在减少计算量的同时保证了稳像精度与范围;进而提出主动运动平稳性和一致性两种度量,利用其反馈调整卡尔曼滤波参数,实现主动运动和抖动的准确分离;最后根据帧间抖动特点选择邻近适合图像进行存储,在节省存储资源的同时达到较好的补偿效果。实验结果表明,该文提出的稳像方法具有较高的稳像精度以及较强的抗干扰能力,可用于实时电子稳像产品。

The classical digital image stabilization system usually suffers from the problems of low precision and hardly real time processing. This paper presents a fast and robust digital image stabilization method. The sub-pixel phase correlation is used for local registration, and the global motion is collaboratively obtained according to the consistency of local motions, which can decrease the calculation while maintain the precision and tolerable stabilization range. The stationary and consistency assessments of intentional movement are proposed, which are used as feedbacks to adjust the parameter online for Kalman filter. Finally, according to the shaking character of frames, the adjacent frames are stored selectively to save storage resources and better compensation effect is achieved. Experiments show that the proposed method is accurate and robust, and can be used for real-time digital image stabilization products.