GPU Accelerated FFT-Based Registration of Hyperspectral Scenes

Abstract

Registration is a fundamental previous task in many applications of hyperspectrometry. Most of the algorithms developed are designed to work with RGB images and ignore the execution time. This paper presents a phase correlation algorithm on GPU to register two remote sensing hyperspectral images. The proposed algorithm is based on principal component analysis, multilayer fractional Fourier transform, combination of log-polar maps, and peak processing. It is fully developed in CUDA for NVIDIA GPUs. Different techniques such as the efficient use of the memory hierarchy, the use of CUDA libraries and the maximization of the occupancy have been applied to reach the best performance on GPU. The algorithm is robust achieving speedups in GPU of up to 240.6×.

Alignment of Hyperspectral Images Using KAZE Features

“…Third, regarding the scale space, it is built following a pyramidal scheme as in SIFT ( Figure 7, lines 5-18), i.e., the image is subsampled for each octave using a bilinear interpolation. Moreover, HSI-KAZE uses FED for the discretization of the diffusion equations such as in A-KAZE ( Figure 7, lines [8][9][10][11][12][13][14][15][16][17]. FED schemes are characterized by their faster computation, ease of implementation, and higher accuracy than AOS approaches.…”

“…Our proposal is compared with other methods in the literature. In particular, it is compared with two methods based on the Fourier transform, the Fourier-Mellin invariant symmetric phase-only matched filtering (FMI-SPOMF) [39] and the HYperspectral Fourier-Mellin algorithm (HYFM) [9], and on the other hand, with the original KAZE [12] and A-KAZE methods [17].…”

“…For KAZE and A-KAZE, the band of highest entropy is selected to perform the registration. On the other hand, FMI-SPOMF uses the first PCA component to perform the registration, while HYFM uses the first 8 PCA components [9]. The random sample consensus (RANSAC) algorithm was used in KAZE and A-KAZE in order to compute the registration parameters from the keypoints and is available in the OpenCV Library.…”

“…They can be classified into two categories according to their nature [6]: area-based and feature-based methods. In the first group, the methods based on correlation between images [7], mutual information (MI) [8], and Fourier transform [9] stand out. These methods work directly with image intensity unlike feature-based methods which seek to detect distinctive features in objects or interest points at a higher level.…”

Accelerating SAR Image Registration Using Swarm-Intelligent GPU Parallelization

“…The use of GPU accelerates the registration based on scale-invariant feature transform (SIFT) by 5.5 times [26]. For hyperspectral images, FFTbased registration has also been completely transplanted to GPU [27]. These methods focus on porting existing methods to the GPU side as completely as possible.…”