الفهرس 
ACKNOWLEDGMENTSOnly 14 pages are availabe for public view
 |  | from | 156 |  |  |
| | | from | 156 | | |
Abstract
The high and accelerating rate of the urban changes and extensions in constructing new cities and buildings in Egypt, calls for an efficient and fast technique for mapping and updating the Geographical Information System (GIS) database. Remotely sensed images can provide a large amount of data for building extraction that has a significant role in a variety of applications such as; urban planning, change detection, disaster monitoring, emergency services, transportation, urban monitoring, and map updating, etc. Automatic detection of buildings from satellite images is a difficult task for many reasons. Building roofs are composed of different materials and some of them are made of non-bright materials. The diverse materials used for the rooftops may have a similar texture and spectral properties with the surrounding objects. Buildings may be occluded with trees or shadows. The present work aims to find the most suitable building extraction approach that can be applied in the Egyptian environment for map updating.
To carry out this study, a GeoEye satellite image acquired in 2014 and a topographic map with a large scale at 1:5000 are used. Buildings extraction from satellite images in this work passes through five steps. In the first step, the GeoEye satellite image is pre-processed using geometric correction and data fusion to obtain a geo-referenced pan-sharpened image. In the second step, a comparison is carried out between two classification techniques on four study areas with different planning degree. The techniques are the traditional pixel-based image analysis and the object-oriented image analysis. Classification accuracy assessment has been calculated for each technique in the third step. In the fourth step, the best classification method result for the good-planned area is refined based on shadow, context information, shape information, Digital Surface
Abstract
ii
Model (DSM) data, and Normalized Digital Vegetation Index (NDVI). The last step is map updating and evaluating of building extraction rate. This step is carried out by converting the refined image into vector format, then comparing it with a map at a scale of 1:5000.
The pre-processing and the pixel-based classification are carried out using ERDAS IMAGINE 2013 software, while the object-based classification and the refinement process are carried out using eCognition Developer 9.0 software. MATLAB R2016a software is used to carry out the vectorization process. The results of this work show that the object-based analysis gives more accurate results than the pixel-based one. Also, it has been noted that maximum likelihood as a pixel-based classifier and Support Vector Machine (SVM) as an object-based classifier achieve the highest overall accuracy than other classification methods. For the Egyptian environment, as the planning degree of the area increases, the classification results become more accurate. The geometric accuracy and the information content of very high-resolution satellite images show the capability of updating 1:5000 building maps.
Keywords: Building extraction, Pixel-based, Object-based, Classification, Accuracy assessment, Refinement, Vectorization, Map updating.