基于先验概率模型的自适应背景图像分割算法  

New adaptive algorithm for segmenting image background based on prior probability model

作　　者：陈伟[1] 唐红光[2] 

机构地区：[1]广东药学院医药信息工程学院,广州510006 [2]海南大学三亚学院理工分院,海南三亚572022

出　　处：《计算机工程与应用》2009年第10期176-178,181,共4页Computer Engineering and Applications

摘　　要：在利用背景消减的视频图像分割过程中,背景模型假定为高斯模型下判断像素点是否为背景点一般采用规则。当模型中方差参数的值较大时,与背景相近的前景被误分割为背景的误差就较大。针对这一问题,提出了一种基于先验概率模型与距离因子对背景进行分割的算法,该算法判定当前帧像素点为背景的概率由其先验概率及该像素点在上一帧分割结果中与前景点的距离决定。实验结果表明,与判定规则相比,该方法在背景变化范围较大的情况下,可以减少前景点被误分割为背景点的误差。In the process of image segmentation in video sequences using background subtraction,when the distribution of background pixels assumed as Gaussian distribution,method using rule is always used to judge whether a pixel belong to be background or not.The error of that the pixel of foreground it similar to the pixel of background will be segmented as background will increase if the value of variance of the model is greater than normal.To overcome this drawback,a novel algorithm is proposed to segment background based on prior probability and distance factor,the algorithm calculates the probability of a pixel belong to be background depend on its prior probability and the distance to the foreground at the last frame segmentation.Experimental results show that the proposed algorithm can reduce the error of the pixels of foreground to be segmented as pixels of background compared with rule when changes in background occur quickly.

关 键 词：自适应分割 运动检测 高斯模型 先验概率 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]