Exploring the Registration of Remote Sensing Images using HSI-KAZE in Graphical Units

Abstract

Registration of hyperspectral remote sensing images is a common task in many image processing applications such as land use classification, environmental monitoring and change detection. The images to be registered present differences as a consequence of being obtained from different points of view, differences in the number of spectral bands captured by the sensors, in illumination and intensity, and also changes in the objects present in the images, among others. Feature-based methods as HSI-KAZE are more efficient at registering than area-based methods when the images are very rich in geometrical details, as it is the case for remote sensing images. But they present, nevertheless, the problem of being computationally more costly because the number of distinctive points to be calculated for these images is high. HSI-KAZE is a method to register hyperspectral remote sensing images based on KAZE features but considering the spectral information. In this work, a robust and efficient implementation of this method on programmable GPUs is presented