计算机工程与应用 ›› 2013, Vol. 49 ›› Issue (19): 127-130.

• 图形图像处理 • 上一篇    下一篇

高阶马尔可夫随机场下目标识别模型的建立

王彩凤1,马  超2,廖福成1   

  1. 1.北京科技大学,北京 100083
    2.首都经贸大学,北京 100070
  • 出版日期:2013-10-01 发布日期:2015-04-20

Object recognition modeling based on high order Markov Random Field

WANG Caifeng1, MA Chao2, LIAO Fucheng1   

  1. 1.University of Science and Technology Beijing, Beijing 100083, China
    2.Capital University of Economics and Business, Beijing 100070, China
  • Online:2013-10-01 Published:2015-04-20

摘要: 目标识别是指一个特殊目标(或一种类型的目标)从其他目标(或其他类型的目标)中被区分出来的过程。给出了高阶马尔可夫随机场下的区域邻域系统定义;通过贝叶斯分析,构建了基于协方差矩阵描述子刻画的图像区域度量的先验模型和似然模型;应用随机算法得到极大后验估计,求得目标所在位置和角度;再通过以目标所在位置为中心,获得多个随机矩形;最终以覆盖范围最大者为所寻找的目标区域。通过Matlab仿真实验,对道路中的斑马线进行模拟识别。实验结果表明,可以达到在大区域中识别出既定目标的目的。

关键词: 高阶马尔可夫随机场, 目标识别, 贝叶斯分析, 标值点过程, 马氏链蒙特卡罗方法

Abstract: Object recognition is to single out a special target from one or some observed images. This paper defines regional neighborhood system based on higher order Markov Random Field(MRF). By Bayesian analysis, a covariance matrix descriptor for regional images is used to model the priori and the likelihood. After modeling, a maximum posteriori estimation can be obtained by Metropolis algorithm. So the central point’s location and angle of the special target in the observed images can be determined. By generating many random rectangles along this central point, it can ultimately recognize the greatest coverage of the special target in the observed images. The Zebra Crossing Line is identified by programming in Matlab 7.6.0. Experimental results show that the special target can be identified from a large image observed.

Key words: high order Markov Random Field(MRF), object recognition, Bayesian analysis, marked point process, Markov chain Monte Carlo