Video human segmentation based on multiple-cue integration

作者：

Highlights：

• Local color segmentation model by training competing 1-SVM and shape priors of video human objects based on motion estimation are integrated into the image segmentation framework.

• The weights of different energy terms in the segmentation model are adaptively adjusted according to relevant feedback cue.

• Our method improves the automatic segmentation accuracy for the subsequent frames because of significantly reducing the adverse impacts on the segmentation for the subsequent frames caused by segmentation error of the current frame.

摘要

•Local color segmentation model by training competing 1-SVM and shape priors of video human objects based on motion estimation are integrated into the image segmentation framework.•The weights of different energy terms in the segmentation model are adaptively adjusted according to relevant feedback cue.•Our method improves the automatic segmentation accuracy for the subsequent frames because of significantly reducing the adverse impacts on the segmentation for the subsequent frames caused by segmentation error of the current frame.

论文关键词：Video segmentation,C-1SVMs,Shape prior,Graph cuts,Multiple-cue integration

论文评审过程：Received 21 May 2014, Revised 26 August 2014, Accepted 2 October 2014, Available online 18 October 2014.

论文官网地址：https://doi.org/10.1016/j.image.2014.10.001