一种基于分水岭和区域合并的视频对象分割算法  

Video Object Segmentation Algorithm Based on Watershed and Region Merging

作　　者：邓庆林[1] 刘海华[1] 吴良健[1] 

机构地区：[1]中南民族大学生物医学工程学院,武汉430074

出　　处：《现代科学仪器》2010年第1期34-38,共5页Modern Scientific Instruments

基　　金：国家自然科学基金资助项目(60972158)

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,并且改善分水岭算法存在的过分分割问题,本文提出了一种基于时空分割融合的视频分割算法。时间分割采用图像差分算法,其关键的阈值选取通过直方图分析得到。空间分割时,先采用抗噪声能力更强的多尺度形态学算子求取图像梯度,然后进行分水岭变换,最后进行区域合并,它可以有效克服噪声的影响并有效地减少过分割。仿真结果表明,此算法对于改善过分割现象有明显效果,并且对分割程度具备可控能力,在适应不同场景需求方面存在一定的应用价值。In order to obtain integrated and consistent segmentation of motion video objects,and improve the problem of excessive partition in watershed algorithm,this article presents a video partition algorithm which based on merged space and time partition.Time partition adopted difference partition algorithm of image,the selection of crucial threshold was got from histogram analysis.In space partition,at fi rst we adopted multi-scale morphological operators which have stronger anti-noise ability to seek the gradient of the image,then processed watershed transform,last did region merging.It can overcome the influence of noise and reduce excessive partition effectively.Simulation results showed that this algorithm has a noticeable effect in improving the phenomenon of excessive partition and has the capacity of controlling the degree of partition.It has a certain value in the aspect of adapting to requirements of different scenes.

关 键 词：时空分割 分水岭算法 多尺度形态学梯度 区域合并 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]