Paweł Kowaleczko
supervisor: Przemysław Rokita
Super-resolution reconstruction (SRR) of natural colour images is one of the most popular research topics in the area of computer vision. In recent years most developed solutions are based on deep neural networks. The generalization of natural colour images SRR is hyperspectral images (HSI) SRR. In this case, each image contains many narrow spectral bands, which usually results (due to the limitations of sensors) in lower spatial resolution when compared to natural images. Most of the natural images SRR methods can also be applied to the HSI SRR, however, better results are usually obtained when the method is designed specifically for the HSI. HSI SRR methods can be grouped into four categories - single image super-resolution (SISR), multi-image super-resolution (MISR), multispectral and hyperspectral image fusion (MSI/HSI fusion) and pansharpening. In SISR a single HSI is used for reconstruction, in contrast to MISR, in which multiple HSIs representing the same object are used for this task. MSI/HSI fusion fuses low resolution hyperspectral image with high resolution multispectral (e. g. natural colour) image. A special case of MSI/HSI fusion, in which the MSI is replaced by high resolution single channel image, is called pansharpening. In this work, the results of multiple SRR methods applied for satellite imagery are presented.