基于光流场的图像分割  

A Better Image Segmentation Method Based on Optical Flow

作　　者：焦春林 高满屯[1] 曹秋萍[1] 陈淑梅 

机构地区：[1]西北工业大学机电学院,陕西西安710072 [2]九五九七二部队,甘肃酒泉735018

出　　处：《西北工业大学学报》2006年第2期265-269,共5页Journal of Northwestern Polytechnical University

基　　金：航空基础科学基金(03553062)资助

摘　　要：基于光流场的图像分割通常用到3个光流模型:仿射变换模型、平面光流模型和二次光流模型,其中仿射模型和平面模型是二次模型的特例。当分别用这3个模型拟合同一光流场时,仿射模型的拟合残差平方和最大,二次模型最小,平面模型居中。但是回归分析的理论表明二次模型并不一定总是最优的。应用回归分析有关待估参数的假设检验准则,可以在这3个模型中选择1个相对最优模型并用于拟合光流场,实现图像分割。算例表明,应用回归分析的假设检验准则选择出最优光流模型,用于图像分割可以取得较好的分割结果。In image segmentation based on optical flow, there are three commonly used optical flow models ; affine transformation model, plane optical flow model and quadratic optical flow model ; the affine model and the plane model are actually special cases of the quadratic model. The first two models are usually employed by past researchers. Ref. 4 presents a segmentation method, which consists of three stages that utilize the three models in turn. When we use the three models to fit a same optical flow field, the residual sum of squares of the affine model is the biggest, the plane model＇s is the second biggest and the quadratic model＇s is the smallest. Nevertheless, the theory of regression analysis tells us that the quadratic model is not always correct. Ref. 5 gives the criterion based on regression analysis, which we employ to select which of the three models is the best one. Lastly, we give two numerical examples, which show preliminarily that our method, which utilizes the criterion given by Ref. 5, does give better results for image segmentation based on optical flow.

关 键 词：图像分割 光流场 回归分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]