high performance computing on graphics processing units: hgpu.org

We present a method for estimating the point spread function (PSF) of solar surface images acquired from ground telescopes and degraded by atmosphere. The estimation is done by retrieving the wavefront phase using a set of short exposures, the speckle reconstruction of the observed object and a PSF model parametrized by Zernike polynomials. Estimates of the wavefront phase and PSF are computed by minimizing an error function with a cooperative particle swarm optimization method, implemented in OpenCL to take advantage of highly parallel GPUs. A calibration method is presented to adjust the algorithm parameters for low cost results, providing solid estimations for either low frequency and high frequency images. Results show that the method has a fast convergence and is robust to noise degradation. Experiments run on a NVidia Tesla C2050 were able to compute 100 PSFs with 50 Zernike polynomials in ~36 minutes. The algorithm is also not affected by the number of Zernikes used, i.e., execution time increased only 17% when the number of Zernike coeffcients increased tenfold, from 50 to 500.