My Thesis
The hypothesis of my thesis is that raw satellite image data can be integrated into an existing geo-dataset by registering the raw data into a reference coordinate system and applying precision correction using wavelet feature detection. The objectives of my thesis are to propose a conceptual design for such an image processing system, to find a suitable map projection for on-board image geo-referencing and to obtain sub-pixel image-to-image registration using wavelet feature detection.
Onboard Geo-referencing:
A newly acquired satellite image needs to be registered in a reference coordinate system so that orientation sensitive matching algorithms can be used to match features in the raw image with features in a reference image.
Traditionally image registration is done manually by an operator using a GIS software package such as Arcview or Grass. Automatic image registration will be an important part of an onboard image geo-referencing system. There are a couple of methods by which to register an image. The method that we are focusing on is image to image registration. With this method the raw image is registered to a reference image so that all image features are exactly correlated and all image pixels can be said to be "in register". Figure 1 illustrates the registration process.
Figure 1. Flow diagram illustrating the geo-referencing process.
Band registration:
I have also successfully applied the wavelet based image registration algorithm to a band registration problem that was encountered by the South African Agricultural Research Council (ARC).
Figure 2. An ARCEagle image before correction showing band misalignment.
Figure 3. The same ARCEagle image after inter-band registration has been performed. The band-misalignment is no longer visible.
Inter-band registration was achieved at sub-pixel accuracy using a second degree polynomial distortion model. A calibrated camera distortion model should give better results than a general polynomial model.
This page was last updated on Sunday, 07 August 2005.