Use of fourier transformations and wavelets for satellite image processing

Abstract

Image processing mainly comprises of operations based on either the pixel values or the operations involving the values of a certain neighbourhood around the pixel, in latter the computations become extensive and the values are affected by the surroundings spatial features. However another approach that converts the image into its frequency components allows visualization and computations based upon spatial frequency rather than the spatial contingency. This approach yields some useful possibilities along with the apparent difficulties in visualizing the frequency domain. Fourier analysis and wavelet transforms are available in most image processing software allowing to be applied in specified ways e.g. Fourier filtering, wavelet compressions, wavelet resolution merges. In this study the capabilities of important image processing software in this context is discussed along with a few implementations of the techniques to de-stripe data, compress data, merge data, visualize data. The implementation is found to be rather difficult to customize for a particular application and the available options in software are somewhat crudely applicable to only specific problems. The applicability of the techniques is also mostly based on hit and trial method especially in case of Fourier filtering unless the exact nature of the solution is predetermined. To cope for this constraint of pre-defined applicability, the customizability of the wavelets using different kinds of wavelets is assessed and Matlab's customization options are used in this study.