计算机科学与探索 ›› 2019, Vol. 13 ›› Issue (2): 285-299.DOI: 10.3778/j.issn.1673-9418.1710063

• 人工智能与模式识别 • 上一篇    下一篇

梯度约束SLIC的快速视频目标提取方法

桂  彦1,2+,汤  问1,2,曾  光1,2   

  1. 1. 长沙理工大学 计算机与通信工程学院,长沙 410114
    2. 长沙理工大学 综合交通运输大数据智能处理湖南省重点实验室,长沙 410114
  • 出版日期:2019-02-01 发布日期:2019-01-25

Gradient-Constrained SLIC Based Fast Video Object Segmentation

GUI Yan1,2+, TANG Wen1,2, ZENG Guang1,2   

  1. 1. School of Computer & Communication Engineering, Changsha University of Science & Technology, Changsha 410114, China
    2. Hunan Provincial Key Laboratory of Intelligent Processing of Big Data on Transportation,Changsha University of Science & Technology, Changsha 410114, China
  • Online:2019-02-01 Published:2019-01-25

摘要: 提出了一种基于梯度约束SLIC(simple linear iterative clustering)的快速视频目标提取方法,允许在关键视频帧上提供少量用户交互下,该方法能够快速并精确地提取复杂视频片段中的视频目标。首先,采用梯度约束的SLIC算法对视频片段进行预处理,有效降低待处理的视频数据量;其次,以预处理生成的超像素为结点构建三维无向图,在此基础上定义能量函数,并结合外观特征与运动特征建立鲁棒的相似外观度量机制;最后,采用最大流/最小割算法最小化能量函数以得到三维无向图的最优划分,从而最终实现视频目标提取。实验结果表明,该方法在处理包含复杂场景的视频片段时能够获得理想的视频目标提取结果,且时间效率相比现有视频目标提取方法明显提高。

关键词: 视频目标提取, 超像素分割, 超像素, 图割优化

Abstract: This paper proposes a fast interactive video object extraction method based on gradient-constrained SLIC(simple linear iterative clustering) algorithm, which can quickly and accurately extract the video object in a complex video with less user interaction. Firstly, each video frame can be segmented into individual superpixels by using the gradient-constrained SLIC algorithm, which effectively reduces the complexity of the subsequent processing. Secondly, a three-dimensional undirected graph is constructed with the generated superpixels, and then a novel graph-cut based energy function is redefined with the robust measurement distance on appearance information and motion features. Finally, video object can be extracted when minimizing the energy function by using the maximum flow/min cut algorithm. The experiments show that this method can achieve high quality video object extraction even dealing with high-definition video fragments with complex scenes, and the time efficiency of video object extraction is obviously improved compared with the existing video object extraction methods.

Key words: video object extraction, superpixel segmentation, superpixel, graphcuts optimization