Semi-automatic registration of satellite images using digital region extraction.

number: 
295
English
department: 
Degree: 
Imprint: 
Physics
Author: 
Atheel M.J. Al-Shalchi
Supervisor: 
Dr.Rehab H.Alwan
year: 
1998
Abstract:

Registration in remote sensing means the rectification of an image to another, i.e., to remove the geometric distortion exists in one of the images considering one of them as a reference. Normally, registration between two digital images depends on the selection of clear points which are called control points The number of control points will limit the polynomial order of the transformation used in the registration process. The selection of these control points usually taken visually, in this present work an attempt is done to automate the control points selection and then registration of the two given shifted digital images was done. Regioning method is used. It includes global mean and local mean thresholding and the suggested Fourier Segmentation method. They are implemented to extract regions, then edge detection and linking methods arc applied to obtain a closed boundaries of regions. Center of gravity extraction is made to determine the control points and their positions. Matching method is applied using Fourier similarity method. Once the control points pairs are matched, a geometric transformation equations are applied to get the final registered image. The program ot semi-automatic registration was pairs of satellite images containing distortions of shift, rotation and scale. The RMS error of registration of those pairs was 2, 4 and less than one pixel respectively. The automation and accuracy achieved in the proposed registration process gave very good results.