基于核聚类的H .264压缩域的运动对象分割方法

摘要

In this paper,a moving object extraction method in H. 264 compressed domain based on kernel clustering is proposed. First,the motion vector is normalized. Second, the median filter using Average Euclidean Distance is used to process noise for motion vector. Third,the clustering center is initialized by the subtractive clustering and the use the fuzzy clustering to classify the data. Finally,the ex-tracted condition is produced by the valid function and the moving object is got adaptively by the divided condition. The Hall experiment results show that the method can get the moving object very well.%提出了一种基于核聚类的H.264压缩域的运动对象分割的方法。首先对运动矢量进行归一化处理，其次又引入了平均欧式距离的中值滤波的方法对运动矢量进行滤波去噪，再次利用减法聚类初始化聚类中心并且使用引入了核函数的模糊聚类进行聚类，最后将有效函数的判断结果作为分割条件，从而达到自适应分割出压缩域中的运动对象的目的。本方法对Hall视频序列进行了实验。实验证明，通过该方法可以较好地分割出视频序列中的运动对象。