| Summary: Due to some technical and non-technical reasons, it may not always be possible to obtain remotely-sensed images of high spatial resolution. Pansharpening offers a robust solution for this problem. Pansharpening aims to transfer the spatial detail content of a high spatial resolution panchromatic (PAN) image into a lower spatial resolution multispectral (MS) image, producing an MS image of the same spatial detail quality as the PAN image. A wide range of pansharpening methods have been proposed so far. Of all, the component substitution (CS)-based pansharpening methods draw attention owing to their simplicity and ability to sharpen images. However, the CS-based methods tend to distort the colour features of the input MS images, due to the inconsistencies between the input PAN image and the intensity component computed from the input MS bands. A wide variety of approaches have been developed to estimate the contributions of the input MS bands on the intensity component to minimize the colour distortion. The previous attempts revealed the fact that improving the colour quality causes spatial distortion to a certain degree, which means that more robust solutions are needed to find the best balance between the spectral and spatial quality offered by the CS-based pansharpening methods. Hence, this thesis, for the first time in the literature, proposed to use the symbiotic organisms search (SOS) algorithm, one of the most powerful metaheuristic optimization algorithms, to estimate a weight for each input MS band in order to optimize the intensity components used by the CS-based synthetic variable ratio (SVR) method and a hybrid method that includes both the intensity-hue-saturation (IHS) and discrete wavelet transform (DWT) methods. This thesis also proposed to use the multi-objective version of the SOS algorithm (MOSOS) to find the best compromise between the spatial and spectral fidelity offered by the SVR and IHS-DWT methods. Using the MOSOS algorithm with these methods also enabled the production of pansharpened images of required spectral or spatial quality. The performance of the proposed SOS-SVR, SOS-IHS-DWT, MOSOS-SVR and MOSOS-IHS-DWT methods were qualitatively and quantitatively compared against some of the very popular pansharpening methods in four test sites with different characteristics. The results showed that the SOS-SVR and SOS-IHS-DWT methods presented a superior colour preservation and spatial detail transfer performance, compared to the other methods used. It was also concluded that the proposed MOSOS-SVR and MOSOS-IHS-DWT methods succeeded in finding the best balance between the spectral and spatial quality; and in producing images of required spectral and spatial quality. These methods were also found to be very successful in producing images of extreme colour and spatial detail quality. The evaluations also revealed that the proposed methods did not show a very good performance on single-sensor input images, but also on multi-sensor input images. They were also found to be able to handle low-contrast images and input images with a high spatial resolution ratio difference. The SOS algorithm does not use any algorithm-specific parameters that may affect the pansharpening performance, which makes it very appropriate for pansharpening process, avoiding analyst intervention. |
