一种基于像素梯度信息的背景减除法  

Background Subtraction Based on Local Gradient Feature

作　　者：张小骏[1] 刘志镜[1] 陈昆[1] 

机构地区：[1]西安电子科技大学计算机学院,西安710071

出　　处：《计算机科学》2015年第8期300-304,共5页Computer Science

摘　　要：讨论了背景模型的更新参数与模型精度的关系。通过精确的梯度背景模型值间接估计当前帧中背景像素理论上的期望梯度值。以高斯模型为基础,将当前帧背景像素的实际梯度值与其理论上的期望值进行比较,计算偏差概率,以此为基础,形成不依赖于局部纹理的梯度特征的相似性度量方法。再用梯度特征的相似度量化地调整差分图像在各像素点处的二值化阈值,实现像素值信息与梯度信息的融合使用。实验表明,本方法对前景分割有一定的改善效果。The relation between accuracy and the study speed of the background model was discussed. The theoretical gradient expectation was estimated with the accurate gradient background model. Based on Gaussian model, the proba- bility of the deviation between the actual gradient and its expectation was given, leading to a similarity measurement of the gradient feature using no texture message. The similarity was then used to adjust the threshold for binarization of the difference image, which means the fusing use of grey level message and the gradient message. Experiments show that the proposed method does have some improvement on foreground segmentation.

关 键 词：背景减除法 噪声 高斯模型 梯度特征 相似度 预计算 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]