基于双向融合机制的建筑目标检测方法  

Approach to building extraction based on two-way fusion mechanism

作　　者：金泰松[1] 汪哲慎[2] 叶聪颖[2] 李翠华[2] 

机构地区：[1]北京理工大学计算机科学与技术学院计算机科学工程系,北京100081 [2]厦门大学信息科学与技术学院计算机科学系,福建厦门361005

出　　处：《计算机应用研究》2008年第2期455-457,共3页Application Research of Computers

基　　金：国家创新研究群体资助项目(60024301);国家自然科学基金资助项目(60175008);厦门大学“985”二期信息创新平台资助项目

摘　　要：模拟人类视觉中有意识主动寻找与无意识被动受吸引相交互的视觉过程,提出了一种基于双向融合机制的建筑目标检测方法。该方法综合了基于自底向上数据驱动的视觉显著性模式和自顶向下基于环境感知的目标搜索模式检测图像中的建筑目标区域。利用拍摄的自然图像进行实验表明:该方法能较好地检测出图像中的建筑目标,具有处理速度快、准确性高的特点,能够满足处理复杂场景图像的实时性要求。This paper proposed an approach called two-way fusion mechanism, which detected the interested building in the image, it imitated the detection process of human vision, in which searching subjectively and being attracted objectively happened at the same time, Through the bottom-up methods which utilized visual salience model and the top-down methods based on environmental perception technology, the regions where building exist were marked in the image. Experiment on natural images validates that this mechanism has competitive accuracy in building detection in complex scene and favorable speed, and satisfies the requirement of processing the natural image.

关 键 词：双向融合 建筑 显著性 竖直线 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]