摘要
为了从视频序列中分割出完整的、一致的运动视频对象,该文使用基于模糊聚类的分割算法获得组成对象边界的像素,从而提取对象。该算法首先使用了当前帧以及之前一些帧的图像信息计算其在小波域中不同子带的运动特征,并根据这些运动特征构造了低分辨率图像的运动特征矢量集;然后,使用模糊C-均值聚类算法分离出图像中发生显著变化的像素,以此代替帧间差图像,并利用传统的变化检测方法获得对象变化检测模型,从而提取对象;同时,使用相继两帧之间的平均绝对差值大小确定计算当前帧运动特征所需帧的数量,保证提取视频对象的精确性。实验结果证明该方法对于分割各种图像序列中的视频对象是有效的。
In order to obtain Integrated and consistent segmentation of motion video objects, the segmentation algorithm based on fuzzy clustering is used to obtain pixels that constitute boundaries of motion object and extract video objects in the sequence sequentially. The motion properties of the current frame in each wavelet sub-band first are calculated using information of current frame and some frames before it in this algorithm. The set of motion eigenvectors is constructed with these properties. Then, significant change pixels are separated by fuzzy C-mean clustering algorithm based on these motion eigenvectors of low-resolution image. The change detection mash of motion object is obtained with significant change image, instead of frame difference, by conventional method of change detection and video objects are extracted. At the same time, the mean absolute different between consecutive two frames is used to determine number of frames that are used for properties calculation. It ensures accuracy of video objects obtained,. The experimental results demonstrate the algorithm effective.
出处
《电子与信息学报》
EI
CSCD
北大核心
2006年第9期1689-1692,共4页
Journal of Electronics & Information Technology
基金
国家自然科学基金(30370393)
国家民委自然科学基金(MZZ04004)资助课题
关键词
视频对象
对象分割
模糊C-均值聚类
Video object, Object segmentation, Fuzzy C-mean clustering
