一种基于确信度的双向视频分割方法

提出了一种将前景物体从一个视频序列中分割出来的方法.提出了确信度模型,该模型由局部颜色配置信息求得.该算法首先对视频所有帧做了分水岭预处理,然后对关键帧做了图分割处理.接下来经由双向过程计算确信度.具体为先通过正向过程计算确信度,再通过逆向过程并辅以光流算法对小部分确信度进行修正,最后根据确信度进行前背景标注.该确信度模型对前背景较为相似的视频有良好的分割效果,并且由于通过双向过程进行分割,对视频中部分遮挡物体的分割也有不错的改善.

Bi-Directional Video Segmentation Algorithm Based on Reliability

In this paper, an algorithm to segment foreground object from a video clip is proposed. The reliability model, which calculated from the local color pattern information, is presented. Firstly, pre-process all the frames of the video clip with the watershed algorithm; then, apply graph-cut on the key frames. Secondly, the reliability model is been computed by bi-directional procedure. Through the positive procedure, the reliability is set. Next, a small portion of the reliabilities is corrected through a reverse procedure assisted by a certain optical flow algorithm.At last, segment the video on the authority of the reliability of each frame. When dealing with the video which has similar foreground and background color, the reliability model shows a good segment result. And bi-directional procedure provides a way to improve segmentation of the occluded object in the video.