中文摘要：

提出了一种新的基于时域定区间记忆补偿的视频对象分割算法。首先,使用对称帧帧差累计法及帧差图像的四阶矩检测出初始运动变化区域;然后,对检测出的初始运动变化区域通过时域定区间记忆补偿法进行补偿,并进一步整合形成全局运动记忆母板,在空域使用Sobel边缘检测算子较为精确地检测得到当前帧中所有边缘;最后,进行时空融合,从而提取出完整精细的运动对象轮廓并通过填充得到运动对象模板。实验证明了本文算法的正确性和快速性。

英文摘要：

A new video object segmentation algorithm based on temporal memory compensation is proposed.First,the cumulation of inter-frame differences of symmetrical frame images and the fourth-order statistical moment of inter-frame images differences are calculated to obtain the initial image movement variation template;Second,the initial image movement variation template can be compensated by the temporal fixed time interval memory compensation.And then,the compensated image movement variation template of every frame in the video sequence can be integrated to form the global movement memory motherboard.In the spatial domain,Sobel edge detection operator is used to detect the accurate boundary of all the objects in current frame;Third,the temporal and spatial information are integrated to extract the complete and precise boundary of moving object,and then the filling is fulfilled to extract the final moving object template in video sequence.Experimental results indicate the correctness and validity of the proposed video object segmentation algorithm.