A Building Extraction Approach Combining Mathematical Morphology With Spectral-Shadow-Shape Constraints For High Spatial Remote Sensing Imagery

With the development of remote sensing satellite sensor,the spatial resolution of remote sensing images has increased from medium resolution and low resolution to high resolution and even very high resolution.High resolution remote sensing images provide rich information about ground targets in detail,making it possible for image interpretation in a smaller scale.The precise detection of buildings is one of the key information sources for urban planning,change detection and population estimation.Manual interpretation is the popular method for extracting buildings,which does not satisfy the requirement of the fast interpretation for remote sensing images.The supervised classification methods are subject to trivial collection of training samples and process of machine learning.Hence,this paper propose a more automatic building extraction approach,which takes morphological features,spectral features,contextual information and shape features into account.Validated on a series of large test images obtained by the widely used commercial satellite sensors,the experiments show that the accuracies of all the test images are over 80%.The main research work of this thesis is as follows:(1)Building a relationship between the implicit characteristics of buildings(e.g.,brightness,contrast,and size)and the morphological operators(e.g.,top-hat by reconstruction,granulometry,and direction).(2)Morphological operators can only represent the structural information of the image,which neglect the spectral features,contextual information and shape features.This paper proposes the building extraction approach combining morphological operators with spectral features,contextual information and shape features.(3)Analyzing whether the proposed building extraction approach can adapt to various challenging scenes.Although MBI shows satisfied results in urban areas,but it is subject to a large number of false alarms in non-urban regions.Validated on a series of large test images,the experiments show promising results of proposed approach over various areas such as urban,mountainous,rural,and agricultural areas.(4)Comparing with the supervised classification approach and the shadow-based building extraction approaches.