用于场景分类的显著建筑物区域检测  

Salient Building Region Detection for Scene Classification

作　　者：陈硕[1] 于晓升[1] 吴成东[1] 陈东岳[1] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110819

出　　处：《上海交通大学学报》2011年第8期1130-1135,共6页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金资助项目(60874103);青年科学基金资助项目(61005032);中央高校基本科研业务费专项资助项目(N090404001);辽宁省自然科学基金资助项目(20102062)

摘　　要：为了实现城市区域内室外场景的快速分类,提出了一种基于视觉注意力选择机制的显著建筑物区域检测方法.该方法首先通过Gabor算子对图像进行滤波,得到水平方向与垂直方向的联合增强图像,然后利用基于相位傅里叶变换(Phase Fourier Transform,PFT)的显著性映射算法提取视野中的显著建筑物候选区域,最后通过方向直方图判别算法去除干扰目标,并采用CV(Chan-Vese)模型实现显著建筑物区域的分割.该方法在注意力选择机制及建筑物方向特征的基础上,提取具有场景代表性的建筑物区域信息,增强了算法的时效性和场景分析能力,同时实现了区域信息在像素级上的精确检测,并将其应用于美国南加州大学的多类室外场景数据库.实验结果表明,显著建筑物区域检测方法在检测结果、计算速度和鲁棒性等方面均取得了较为满意的效果.In order to achieve outdoor scene classification in urban areas quickly,an effective method for detecting salient building region based on visual attention selection mechanism was proposed.Firstly,eight-directional Gabor filter is used to obtain horizontal-vertical enhanced image.Secondly,salient building candidate regions are extracted through mapping from salient map which is constructed by phase Fourier tranform to the enhanced image.Finally,an orientation histogram discrimination algorithm is used to eliminate interference target,and salient building region detection is completed by CV model segmentation.The method extracts the building regional information which can represent scene based on PFT and orientation features,it effectively improves the real-time of algorithm and the capability of scene analysis,and implements the accurate detection of region in pixel level.The results of experiments with the university of Southern California outdoor scene database show that the method performs well in the detection result,computational speed and robustness.

关 键 词：显著性映射 区域检测 CV模型 Gabor算子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]