GPU-accelerated MRF segmentation algorithm for SAR images

Markov Random Field (MRF) approaches have been widely studied for Synthetic Aperture Radar (SAR) image segmentation, but they have a large computational cost and hence are not widely used in practice. Fortunately parallel algorithms have been documented to enjoy significant speedups when ported to run on a graphics processing units (GPUs) instead of a standard CPU. Presented here is an implementation of graphics processing units in General Purpose Computation (GPGPU) for SAR image segmentation based on the MRF method, using the C-oriented Compute Unified Device Architecture (CUDA) developed by NVIDIA. This experiment with GPGPU shows that the speed of segmentation can be increased by a factor of 10 for large images.

► We present a GPU-accelerated SAR image segmentation method using the CUDA.
► The corresponding speedup is about 10.53.
► Eight cores CPU parallel processing is implement to match our method.
► The speedup on CPU with eight cores is about 7.26.
► Result shows that we run on a low-cost hardware environment but earn a high performance response.