Video retargeting using matching area with seam carving and object detection

摘要

Seam carving is an algorithm for image and video resizing. The main problem facing seam carving is that, seam carving fails to protect important objects in images and video. In this paper, a new video resizing technique is proposed, which combines seam carving with using a matching area technique, this technique is based on matching area based temporal energy adjustment that allows per key frame seam carving to remove the optimal pixels to achieve spatially and temporally continuous resized videos, and object detection to detect the object (regions of interest) based on a single Gaussian background model method. Compared with inter-frame difference method, the single Gaussian background model is much better at object detection because of its better accuracy. After object detection, it is suitable for selecting the objects in the key frames for retaining or removal and by using a matching area between key frames we achieve both spatial and temporal coherence. Experimental results demonstrate that the proposed approach can remove or protect the region of interest and non region of interest in the video frames in an excellent manner and less storage size.