Highly parallel image co-registration techniques using GPUs

Image co-registration is a widely used processing in many application fields, such as medicine, computer vision and remote sensing. In each domain, different techniques may be applied, depending on the nature of the data available and the wanted results. In this paper, we focused on optical and SAR based satellite images, that need to be co-registered for a variety of reasons, from change detection to interferometry, from agriculture monitoring to surveillance and so on. One of the main characteristic of satellite based imagery is the large amount of data that needs to be processed, that usually leads to high running time, thus limiting the kind of processing that can be effectively carried out. To solve this problem, a complete toolchain for image co-registration based on a GPU parallel architecture is presented: the tool flow, the detailed implementation, and experimental results are illustrated, showing the level of performance and accuracy achieved.