Advanced pan sharp techniques with CUDA

Satellite imagery is the taking photographs of a location by using artificial satellites in orbits. However, they are not the images, as we usually understand. The satellites that obtain the images contain some special sensors to capture and record multi-spectral data. These sensors obtain data from a tiny part of the Earth's (or wherever) surface. RGB color space and panchromatic band are some of the data. RGB color space is consist of three components that are red, green and blue also known as primary colors. A color can be obtained by mixing these three base colors. This color space is used by computers, graphic cards and monitors. Panchromatic band is a single-band and a type of grayscale image that has a wide bandwidth. Obtaining a high-resolution multi-spectral image from a satellite is an expensive process. However, getting a high-resolution panchromatic image is relatively cheaper. It is possible to achieve getting a high-resolution multi-spectral image using a low-resolution multi-spectral image and a high-resolution panchromatic image. This process is called as pan sharpening and the output is called as pan sharped image. In this study, the aim is how to make pan sharpening techniques with CUDA -trademark of NVIDIA- which is a parallel computing platform and programming model. Throughout this study three pan sharpening techniques are selected to deal with. These are Simple Mean Algorithm, Brovey Transformation and Hue-Saturation-Value (HSV) Transformation. As a result of studies, implementation of the algorithms on the GPU instead of on the CPU is efficient depending on the dimension of the image and the nature of the algorithm. Twenty images from different sources is used in this study and in every case, a speed-up (CPU-GPU ratio) is acquire. Keywords: Satellite imagery, multi-spectral image, panchromatic band, pan sharpening, pan sharped image, CUDA, Simple Mean, Brovey, HSV