摘要
基于建筑物细部边缘信息在数字航片上的精细纹理表达,首先对原始影像进行边缘检测、主成分分析和基于二阶概率统计的纹理滤波等预处理,然后选择用7像元×7像元的窗口锐化得到Contrast纹理特征的灰度图;采用Contrast灰度图(R)、原始航片(G)、原始航片(B)的波段组合进行假彩色合成,得到基于对比度纹理的假彩色合成影像;最后对假彩色合成影像进行多尺度分割和建筑物提取。以北京市延庆县康庄镇2008年12月数字航摄影像为例,运用上述方法进行村镇建筑物信息提取。结果表明,与运用面向对象的分类方法相比,利用纹理增强提取村镇建筑物信息的方法突出了建筑物边缘,减少了冗余分割对象,解决了建筑物与其阴影相混淆不利于建筑物信息提取的问题;并对特征空间进行优化,避免了模糊分类时纹理特征规则运算缓慢的问题,较完整地提取出了村镇建筑物信息,提高了分类精度。
Based on fine texture expression of the edge of the detailed information in the digital aerial image, the authors first preprocessed the digital aerial image by edge detection, principal component analysis and the texture filter of second - order probability statistics, secondly obtained the gray image of the contrast texture through the sharpening window of 7 × 7, then taking the gray image as an independent band, processed a pseudo color composition with the band combination of contrast ( R), the digital aerial image band (G) and the digital aerial image band (B). Finally, multiple segmentation and building extraction in towns and villages based on the pseudo color images were processed. With the digital aerial image acquired in December 2008 from Kangzhuang Town Yanqing County in Beijing as an example, the authors processed building extraction by using the method mentioned above. Compared with the object - oriented classification method, it not only highlighted the edges of the buildings but also reduced the redundant segmented objects. Besides, it achieved an effective solution of the shadow of the building and its confusing area, optimized the feature space, and improved the classification accuracy.
出处
《国土资源遥感》
CSCD
2010年第4期51-55,共5页
Remote Sensing for Land & Resources
基金
国家科技支撑项目(编号:2006BAJ05A01)和国家科技支撑项目(编号:2008BAK49B01)共同资助
关键词
纹理增强
数字航片
面向对象分类
CONTRAST
假彩色合成
Texture enhancement
Digital aerial image
Object - oriented classification
Contrast
Pseudo color composition