【作者】 陈毅滨； 蔡灿辉；

【Author】 CHEN Yi-bin CAI Can-hui(Institute of Information Science and Engineering,Huaqiao University,Quanzhou 362021)

【机构】 华侨大学信息科学与工程学院；

【摘要】 本文基于立体视频序列的时空特性和左右通道间的视差特性,提出了一种结合对称差分的变化检测和秩空间立体匹配的立体视频对象分割算法。通过三帧的对称差分快速地检测出中间帧的运动对象,并结合视差信息进行前景目标的分割,以进一步提高分割的准确性。为了获得准确的视差信息,将图像从灰度空间变换到秩空间,并把视差分解为全局视差与局部视差之和,在秩空间进行二次立体匹配。对所得的视差场进行迭代阈值分割、二值化后与对称差分所获得的掩膜相与,投影到原始左右图像,获得最终目标。实验结果表明,该方法对于左右视频亮度差异较大、含遮挡区域及背景存在运动的立体视频序列,都有较好的分割效果。更多还原

【Abstract】 Based on temporal-spatial characteristics of stereo video and disparity between left and right views,a new stereo video segmentation algorithm combining symmetrical-displaced-frame-difference-based change detection and rank-transform-based stereo matching is proposed.With symmetrical frame difference mechanism,moving objects in middle frame can be quickly detected.The depth information is also employed to improve the accuracy of segmentation. To obtain the accurate disparity information,stereo image pairs are first mapped from the gray-level space to rank space and a two-level stereo matching scheme is applied to the image pairs in rank space to calculate global disparities and local disparities.Then,by segmenting the depth field using iterative threshold and combining both binary masks,we get the stereo object pairs.Experimental results show that the proposed algorithm can efficiently segment objects in stereo video even there are strong dissimilarities between two views,occluded regions and moving background.更多还原