Abstract
Image/video retargeting is a well-known technique in image processing and computer vision. This technique retargets an image/video to a desired aspect ratio, while simultaneously retain the shape and structure of important objects. Due to the development of display devices, displaying media contents in various devices, such as smart phones, TV, and Tablets, is getting common, and image/video retargeting technique becomes a useful tool. Content-aware image retargeting has been proven to produce satisfying result. However, in video retargeting, both important content and temporal consistency should be preserved. As a result, video retargeting is a more complicate task in comparison with image retargeting. Extending the current image retargeting technique to individually resize video frames may cause jittering artifacts, leading to noticeable discontinuity when playing videos. Many approaches utilize global information of an entire video frame to preserve temporal coherence. However, in implementation, these methods need a number of buffers to save frames, which is comparably expensive. In our method, we propose a resource-limited frame-by-frame algorithm of video retargeting. First, in order to reduce frame buffer of usage, instead of optimizing over the video cube, we perform our optimization in a frame-by frame manner. Second, in the process of resizing current frame, our method only considers the information of previous frame, which is already optimally deformed and streamed. Experiment shows that our proposed method produces promising results compared to previous works, even under limitation of resources.