视觉显著性驱动的运动鱼体视频分割算法  

Video segmentation algorithm for moving fish based on visual significant driven

作　　者：胡春海[1] 万欣[1] 李勇骁 刘斌[1] 赵兴[1] 

机构地区：[1]燕山大学河北省测试计量技术及仪器重点实验室,河北秦皇岛066004 [2]哈尔滨方舟工程设计咨询有限公司,黑龙江哈尔滨150030

出　　处：《燕山大学学报》2017年第1期87-94,共8页Journal of Yanshan University

基　　金：河北省自然科学基金资助项目(F2015203287)

摘　　要：针对养殖监控视频中运动鱼体目标检测问题,提出一种基于视觉显著性计算的运动鱼体视频分割算法.首先,将运动特征通道引入基于图论的显著性计算模型中,对多通道特征进行自动加权,生成全局视觉显著图.其次,利用基于方向梯度直方图的支持向量机分类器对显著区域进行目标确认,得到包含前景目标的感兴趣区域.最后,分割出运动鱼体目标,并将其标注在视频序列中.实验结果表明,所提方法在室内养殖监控数据集上能够很好地剔除水面杂波、反光等复杂背景干扰,实现运动鱼体的视频分割.Aiming at detecting a moving fish target in breeding monitoring video, a kind of moving fish video segmentation algorithm based on visual significant calculation was proposed in this paper.First, on the basis of the graph-based visual saliency （ GBVS） , the motion characteristic channel was introduced. And multi-channel characteristics were automatically weighted, generating a global saliency map.Second, the support vector machine classifier based on histogram of oriented gradients was used to verify the target in significant areas. And then, interest sub-graphs including foreground targets were generated. Finally, the foreground targets which were the moving fishes were segmented out, and then it was marked in a video sequence. The experimental results show that, the proposed method could well remove surface clutter, reflective interference and others complex background on indoor breeding monitoring data set, achieved video segmentation of moving fish body.

关 键 词：图像分割 运动鱼体分割 视觉显著性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]