摘要
提出了一种基于二维网格运动分析与改进形态学滤波空域自动分割策略相结合的视频对象时空分割算法。该算法首先利用高阶统计方法对视频图像的二维网格表示进行运动分析,快速得到前景对象区域,通过后处理有效获得前景对象运动检测掩膜。然后,用一种结合交变序列重建滤波算法和自适应阈值判别算法的改进分水岭分割策略有效获得前景对象的精确边缘。最后,用区域基时空融合算法将时域分割结果和空域分割结果结合起来提取出边缘精细的视频对象。实验结果表明,本算法综合了多种算法的优点,主客观分割效果理想。
A new temporal-spatial video object segmentation algorithm was proposed. This algorithm was based on 2-D mesh-based motion analysis and an improved spatial automatic segmenting strategy of morphological filtering. Firstly, higher order statistics was used for motion analysis of the 2-D mesh representation. After post-processing, relatively fine motion detection mask was quickly obtained. Then a novel watershed segmenting strategy based on alternating sequential filtering by reconstruction and adaptive threshold was applied to get the accurate boundary of the foreground object. Finally, the video object was effectively extracted by fusing the temporal and spatial segmentation via region-based method. Experimental results indicate that the proposed algorithm assimilates the advantages of 2-D mesh-based algorithms and pixel-based algorithms, and achieves satisfactory subject and object performance.
出处
《通信学报》
EI
CSCD
北大核心
2007年第8期76-86,共11页
Journal on Communications
基金
国家自然科学基金资助项目(60672094)
江苏高校高新技术产业发展项目(JH02-076)~~
关键词
视频对象分割
二维网格
交变序列
自适应阈值
分水岭算法
video object segmentation
2-D mesh
alternating sequential
adaptive threshold
watershed algorithm
作者简介
王煜坚(1980-),男,江苏金坛人,东南大学信息科学与工程学院博士生,主要研究方向为视频/图像分割和基于对象的视频编码。
高建坡(1975-),男,山东金乡人,东南大学信息科学与工程学院博士生,主要研究方向为人脸检测、分割与跟踪。
吴镇扬(1949-),男,江苏兴化人,东南大学信息科学与工程学院教授、博士生导师,主要研究方向为多媒体信号处理。