Unsupervised Pixel-level Video Foreground Object Segmentation Via Shortest Path Algorithm

The purpose of the video object segmentation is to separate foreground object from consecutive video frames. Research for object segmentation algorithm has made great progress, there are a lot of well-known algorithms. It has been used in various aspects, such as video retrieval, video analysis and understanding, video summarization and indexing, and has become a hot topic in computer vision field.Traditional methods is based on pixel-level segmentation, which leads to over segmentation. In this paper, we propose an object-level unsupervised method to extract the foreground object in the video. Unsupervised video object segmentation is to automatically segment the foreground object in the video without any prior knowledge. We firstly generate all the object-like regions as the segmentation candidates. Then based on the corresponding map between the successive frames, the video segmentation problem is converted to corresponding graph model, which selects the most corresponding object region from each frame. The shortest path algorithm is explored to get a global optimum solution for this graph. To obtain a better result, we also introduce a global foreground model to restrict the selected candidates. Finally, we utilize the selected candidates to obtain a more precise pixel-level foreground object segmentation. The experimental results on several datasets demonstrate the robustness and effectiveness of our method. Compared with the state-of-the-art object-level methods, our method does not only guarantee the continuity of segmentation result, but also works well even under the cases of fast motion and occlusion.