基于Seam Carving的逐帧优化视频缩放方法

Frame-by-Frame Video Retargeting Method Based on Seam Carving

视频缩放是近年来数字图像处理领域的一项热点问题.针对整个视频进行缩放的方法会带来庞大的内存占用与计算量,导致效率低下、实用性较差;而针对视频中的每一帧进行缩放的方法难以维持视频的时空一致性.为此,基于Seam Carving方法,提出一种逐帧优化的视频缩放方法.首先逐帧读入视频,按照梯度求出当前帧的能量图,并使用高速缓存的置换思想调整能量图;然后根据能量图找出缝;最后使用线性插值的方法删除缝,得到目标大小的帧.实验结果表明,该方法不仅能够在所处理的每一帧中保持图像的重要内容,并且可以维持整个视频的时空一致性,保持较好的视觉效果.

Video retargeting has received considerable attention in image processing research recently. Due to the limited memory and extensive computation, it is difficult to retarget a large-size video. Frame-by- frame retargeting cannot solve this problem because of low quality of spatial-temporal coherence. In this paper, we present a new seam-carving-based video retargeting algorithm. First, video frames are read one by one and energy map per frame are calculated based on the gradient. Then energy map of current frame is adjusted by considering a few previous frames in a cache replacement method. Finally, seam carving based on linear interpolation is performed to resize current frame. We repeat this procedure until all the frames are resized. Experimental results showed that our algorithm not only preserves the important content, but also maintains both spatial and temporal coherence of the resized videos.