Computer Engineering and Applications ›› 2006, Vol. 42 ›› Issue (33): 1-3.

• 博士论坛 •     Next Articles

An Approach to Building Recognition


  1. 北京理工大学计算机系
  • Received:2006-08-07 Revised:1900-01-01 Online:2006-11-21 Published:2006-11-21



  1. 北京理工大学计算机系
  • 通讯作者: 金泰松 jts2002

Abstract: This paper presents a new approach to recognize the object of building in complex scenes. It extracts a set of line segments ,and turns object recognition into maximum a posteriori estimation. The prior knowledge can be incorporated into target identification process by using Bayesian theory. A new distribution is proposed to model the uncertainty of model features based on Gibbs distribution , which captures phenomena such as the fact that unmatched features due to partial occlusion are generally spatially correlated rather than independent. Fluctuations of matched image features are modeled by Gauss distribution. Finally, a score is proposed to evaluate the region of image,where buildings may exist . Experiments on the natural image sets demonstrated that the proposed score to a certain extent can differentiate between the true object and the false object in complex scenes, and the proposed approach yields substantial improvements over the traditional approach based on edge feature on time performance.

Key words: Bayesian, Gibbs distribution, Gauss distribution, Short line segment

摘要: 本文提出了一种在复杂场景图像中识别建筑物目标的方法。该方法提取图像的短线段特征,将目标识别转化为最大后验概率估计问题,并利用贝叶斯理论将建筑物目标先验知识引入到识别过程中。在吉布斯分布基础上,给出了新的分布表征模板特征在匹配过程的不确定性。该分布反映了目标在被遮挡条件下,不匹配的模板特征分布聚集的现象。利用高斯分布表征图像特征的不确定性,最终给出了一个评判分数评估图像中可能存在建筑物目标的区域。实验表明:本文给出的评估分数能在一定程度上区分图像中的虚假目标和真正目标,并且与传统的以边缘特征为图像特征的模板匹配方法相比,本文给出的识别方法在时间性能上有了较大提高。

关键词: 贝叶斯, 吉布斯分布, 高斯分布, 短线段