Search In this Thesis
   Search In this Thesis  
العنوان
Shadow Detection in High-Resolution Satellite Images /
المؤلف
Besheer, Mohamed Ahmed.
هيئة الاعداد
باحث / محمد احمد بشير
مشرف / احمد عبدالحافظ
مناقش / محمود النقراشى عثمان
مناقش / فراج على فراج
الموضوع
Civil Engineering.
تاريخ النشر
2018.
عدد الصفحات
144 p. :
اللغة
الإنجليزية
الدرجة
الدكتوراه
التخصص
الهندسة المدنية والإنشائية
الناشر
تاريخ الإجازة
10/8/2018
مكان الإجازة
جامعة أسيوط - كلية الهندسة - Department of Civil Engineering
الفهرس
Only 14 pages are availabe for public view

from 144

from 144

Abstract

High-resolution satellite images usually contain a significant amount of shadow casted by elevated objects. Shadows and shading in images occur when objects occlude light from a light source. Image information is affected by such shadows. Shadows obscure image information partially and/or totally, that decreases the accuracy of information extraction and change detection. This thesis is trying to find the best possible ways to separate shadow from non-shadow areas and then discuss the possibility of eliminating or reducing the effect of shadows in the high-resolution satellite image.
Invariant color models are commonly used to detect shadows. An index employs near-infrared band in addition to visible bands while generating a C1C2C3 color model is proposed to achieve a modified C3 index. A comparative study is then applied to Hue Saturation Value (HSV) model, the original C1C2C3 color model, and the modified model to validate the developed index. An object-based radiometric enhancement algorithm for shadow based on the linear correlation correction algorithm is also proposed. The digital numbers of shadow pixels are enhanced according to nearby pixels from the same class. The proposed technique, therefore, segments the image into objects considering the distance between objects of the same class as the primary factor in radiometric enhancement process. The performance of the proposed algorithm has been inspected visually and quantitatively in compare with the widely used methods.

The experimental results show the superiority of the proposed approach in shadow detection, with an overall accuracy of 93.9%, 94.4%, and 95.6% while the result for the original C1C2C3 model was 91.6%, 92.9% and 94.2% for studied areas. The proposed radiometric enhancement algorithm also reduces the shadow effect whether in terms of visual improvement or the change in the value of the digital number of each pixel. The overall supervised classification accuracy increased with about 7% in average when applying the proposed enhancement algorithm.