Fast remote sensing image geometric-correction method based on GPU parallel processing

Geometric-correction is used to correct the distortion between image and corresponding ground target, it is one of the most time-consuming steps in remote sensing data processing, thus becoming a bottleneck restricting the effectiveness of remote sensing data. Therefore, a fast geometric-correction method of remote sensing image, based on GPU parallel processing, is proposed in this paper. First, a three-dimensional (3D) linear transformation is introduced to reduce the computational amount of coordinate projection conversion. Then, the GPU parallel mapping is conducted. Furthermore, two performance optimization methods are used to further improve the efficiency. The approach is tested with GF-1 panchromatic image with Intel i7 CPU and NVIDIA Tesla M2070 GPU. The experimental results show that the processing time is only 2.08 s, with a speedup ratio of 142.42, which could meet the requirement of fast geometric-correction for big-data remote sensing images.