Detection and Characterization of Urban Objects from VHR Optical Image Data

The focus of this research is to develop an image analysis tool for building recognition in very high resolution (VHR) satellite image data. This method consists of two stages of analysis of a VHR satellite scene extract covering an urban area. Firstly, an initial image segmentation is carried out, based on a 4 neighbor connectivity similarity threshold and a segment range constraint. Statistical analysis of local difference in the scene is used to initialize the values of these two parameters. This results in accurate and consistent object detection. This will be followed in subsequent work by a second stage, consisting of the extraction and analysis of object features useful for object recognition. The object analysis will be based on morphological, topological and reflectance features of image segments and will attempt to determine to what extent buildings can be distinguished from other objects in an image, based on the statistical distributions of these features. The results presented in this paper use mostly panchromatic data, but the method is applicable to multispectral image data, which clearly allows better characterization of the spectral signatures of image objects.