摘要
使用静态空间特征通常无法得到准确的视频显著性目标对象,提出了超像素梯度流场与元胞自动机融合的视频图像显著性检测方法。首先,使用SLIC方法将视频帧分割成超像素,在超像素级上运用光流梯度和颜色梯度生成一个时空梯度函数,由时空梯度得到新的梯度流场值,将视频中运动信息充分利用起来;其次,在视频帧超像素图像上使用卷积神经网络得到其深度特征,通过元胞自动机使这些深度特征依自定义规则更新出粗略显著图,然后将梯度流场显著图与元胞自动机粗略显著图融合得到最终的显著图;最后,在ViSal数据集上、采用5种评估指标、与现有的4种方法进行对比实验,结果表明本文方法在动态视频图像显著性检测中有好的表现。
Aiming at that the static spatial features can not get accurate saliency target,a video saliency detection algorithm based on the fusion of superpixels gradient flow field and cellular automata is proposed.Firstly,SLIC method is used to segment video frames into superpixels.At the superpixels level,optical flow gradient and color gradient are used to generate a spatiotemporal gradient function,and new gradient flow values are obtained from spatiotemporal gradient,which can make full use of the motion information in the video.Secondly,the depth features of the video frames are obtained using the convolutional neural network.The depth features are updated by the cellular automata according to the custom rules,and then the gradient flow field saliency map is fused with the cellular automata saliency map to get the final saliency map.Finally,the comparative experiments on ViSal data set use five evaluation indexes and four existing methods,the results show that the method has advantages in the dynamic video image saliency detection.
作者
张荣国
贾玉闪
胡静
刘小君
李晓明
ZHANG Rong-guo;JIA Yu-shan;HU Jing;LIU Xiao-jun;LI Xiao-ming(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China;School of Mechanical Engineering,Hefei University of Technology,Hefei 230009,China)
出处
《太原科技大学学报》
2021年第5期341-347,共7页
Journal of Taiyuan University of Science and Technology
基金
国家自然科学基金(51875152)
山西省自然科学基金(201801D121134)
太原科技大学博士科研启动基金(20202057)。
关键词
超像素
梯度流场
元胞自动机
视频图像
superpixels
gradient flow field
cellular automata
video saliency