利用最小Tsallis交叉熵的视频对象分割算法  

Video object segmentation algorithm based on minimum Tsallis cross-entropy

作　　者：程淑红[1] 胡春海[1] 

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004

出　　处：《红外与激光工程》2011年第3期559-563,共5页Infrared and Laser Engineering

基　　金：河北省教育厅自然科学发展计划项目(2006455)

摘　　要：提出了一种基于改进的最小Tsallis交叉熵的视频对象分割算法。在时域中采用帧间变化检测快速区分前景与背景,在空域中使用改进的最小Tsallis交叉熵设置自适应阈值,能更准确地对差分图像进行阈值分割,克服了传统的最小Tsallis交叉熵法阈值化容易失效的问题,然后利用形态学修正和差分交集技术获得精确的视频对象。实验结果表明,该方法能够准确地识别运动目标,对于分辨率为240×352像素,每像素8位量化的单运动目标序列图像,处理目标分割耗时0.475s;对于分辨率为480×640像素,每像素8位量化的多运动目标序列图像,处理目标分割耗时0.518 s,能够实现对运动目标的实时分析,具有较好的实时性、自适应性、鲁棒性。An automatic segmentation method was presented for moving objects in video sequence based on improved minimum Tsallis cross-entropy.In the temporal domain,the foreground and the background were distinguished quickly by using frame difference detection,while in the spatial domain,the adaptive threshold was set up by use of the method of improved minimum Tsallis cross-entropy,which segmented the threshold of frame difference image more accurately,and overcame invalidation problem of threshold of conventional minimum Tsallis cross-entropy method.Then the morphologic modification and the difference intersection method were used to obtain accurate video objects.The experimental results indicate that the method can successfully detect moving objects.For the resolution of 240×352 pixels and 8-bit per pixel to quantify the image sequences of single-moving target,time-used to deal with object segmentation is 0.475 s;for the resolution of 480×640 pixels and 8-bit per pixel to quantify the image sequences of multi-moving target,time-used to deal with object segmentation is 0.518 s.It can meet the demand in real-time detecting and is efficient,real time,adaptive and robust.

关 键 词：MPEG-4 视频对象分割 差分图像 Tsallis交叉熵 数学形态学 

分 类 号：TN919.8[电子电信—通信与信息系统]