CASR-Net: A color-aware super-resolution network for panchromatic image

Abstract

Spatial resolution is the ability to distinguish the spatial details of remote sensing images, and high spatial resolution images are conducive to object recognition and visual interpretation. Spectral resolution is the ability to distinguish the spectral details of the ground objects in remote sensing images, and high spectral resolution images are of great significance to the classification and recognition of objects in remote sensing images. The image super-resolution model is used to enhance the spatial resolution of remote sensing image, but it cannot enhance the spectral resolution, while the image colorization model can increase the number of channels by predicting chromatic channels for the input image, thereby improving spectral resolution. In this paper, a color-aware super-resolution network that combines image colorization and super-resolution ideas is designed to improve the spectral and spatial resolution of panchromatic images. The color-aware super-resolution network mainly contains color-aware block and spatial-aware block, color-aware block is presented to predict color information for panchromatic images to improve the spectral resolution, meanwhile, spatial-aware block is used to restore the texture details for panchromatic images to improve the spatial resolution. The trained color-aware super-resolution network only needs to input panchromatic images to generate images with more spectral information and higher spatial resolution than input images. Extensive experiments demonstrate that our color-aware super-resolution network has a good performance in image colorization and super-resolution, and experimental results show that compare with some existing excellent image colorization methods and super-resolution methods, our method is excellent in objective indicators and visual effects.