Research on GPU-Based Real-Time MTF Compensation Algorithm

MTF compensation is widely used in remote sensing applications for recovering high frequency components of images. However, MTF compensation is very time-consuming and therefore becomes a major bottleneck in the whole image pre-processing process. In order to address the above-mentioned problem, this thesis proposes a novel GPU-based real-time MTF compensation algorithm with the aim of dramatically improving the computation efficiency of MTF compensation. First, the basic approach "tile strategy" for accelerating MTF compensation on GPU is put forward; this strategy divides the whole image into small tiles and compensates these tiles in parallel. Next, detailed performance tuning approaches including shared memory use, three block layers division and loop unroll are introduced. By careful performance tuning, GPU\´s powerful computing capacity can be fully utilized. Finally, Quickbird panchromatic image is taken for experiment, the result proves that the proposed strategy can dramatically accelerate the MTF compensation algorithm and commendably guarantee the same compensation effects with the CPU compensated image.