摘要
为了提高视频前景对象分割的效率,设计了一个融合半监督及无监督算法的视频前景对象自动分割方法。计算视频帧中超像素,建立视频前景模型更新机制用来调节阈值和环境复杂性。将鬼影作为视频背景进行分类,同时更新背景模型,建立前景的统计图谱,对背景的相似度进行了二次判别。检测视频前景对象阴影,在此基础上,建立象素级别的交叉熵损失函数,对候选区域集成。采用融合半监督及无监督算法对视频前景对象自动分割,以此实现融合半监督及无监督算法的视频前景对象自动分割。实验对比结果表明,此次研究的视频前景对象自动分割方法有效提高了分割效率,并提高了分割全局正确率以及分割轮廓精准度,满足了视频前景对象自动分割需求。
In order to improve the efficiency of video foreground object segmentation,an automatic video foreground object segmentation method based on semi supervised and unsupervised algorithm is designed.The super pixels in the video frame are calculated,and the video foreground model updating mechanism is established to adjust the threshold and the complexity of the environment.The ghost image is classified as the video background,and the background model is updated to establish the statistical spectrum of the foreground.The shadow of foreground object is detected,and then a pixel level cross entropy loss function is established to integrate the candidate regions.The fusion of semi supervised and unsupervised algorithm is used to segment video foreground object automatically,so as to realize the fusion of semi supervised and unsupervised algorithm.The experimental results show that the proposed method can effectively improve the efficiency of segmentation,improve the global accuracy of segmentation and the accuracy of segmentation contour,and meet the requirements of automatic segmentation of video foreground objects.
作者
战涛
姚璐
Zhan Tao;Yao Lu(Xi'an Mingde Institute of Technology,Xi'an 710124,China)
出处
《科技通报》
2022年第5期31-35,共5页
Bulletin of Science and Technology
基金
陕西省教育厅2019年度科学研究计划(项目编号:19JK0866)。
关键词
融合半监督
无监督算法
超像素
视频前景
自动分割
fusion semi-supervised
unsupervised algorithm
super pixel
video foreground
automatic segmentation
