基于先验形状的人工目标识别与轮廓恢复  

Artificial Object Recognition and Contour Restoration Based on Prior Shape

作　　者：尹高飞[1] 肖鹏峰[1] 吴桂平[1] 李晖[1] 周立国[2] 

机构地区：[1]南京大学地理信息科学系,南京210093 [2]复旦大学环境科学与工程系,上海200433

出　　处：《遥感信息》2011年第6期8-12,共5页Remote Sensing Information

基　　金：国家高技术研究发展计划(2008AA12Z106);国家自然科学基金(40801166;41001234);国家自然科学基金(40771137);教育部高等学校博士学科点专项科研基金(200802841012)联合资助

摘　　要：以具有典型形状特征的操场为例,探讨了形状特征在人工目标自动识别中的应用。基于形状特征的人工目标识别,一般首先将图像分割为多个基元,然后依据人工目标有较规则形状的特点,以形状特征作为指标进行识别。由于高分辨率遥感图像细节信息丰富,这种方法所提取的目标轮廓往往不完整。本文提出了"图像-基元-目标-轮廓恢复"的识别模式,在目标识别后采用加入形状先验知识的主动轮廓模型对目标进行轮廓恢复。实验结果表明,这种方法可以有效地修正目标提取结果中的轮廓缺失。In this paper,we propose a new method utilizing prior shape to recognize an artificial object in remotely sensed image of high spatial resolution.As the classic object-oriented method,we get image parcels via threshold segmentation,and extract the target by its peculiar shape feature.Further on,we add a post-processing process,called contour revision,to improve the contour accuracy of artificial objects.The contour revision process is accomplished by an active contour model incorporating with prior shape.In order to capture the shape information,the level set method is used to represent the shape.Then principle component analysis is applied to capture the various mode of shape.After that we combine the shape information with active contour model.Finally,the initial extracted contour is evolved based on the active contour model already mentioned.The validity of this method is testified through an experiment of playground（typical artificial object with regular shape） recognition.

关 键 词：目标识别 先验形状 轮廓恢复 主动轮廓模型 高分辨率遥感图像 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]