Berlin 2015 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
BP: Fachverband Biologische Physik
BP 13: Posters: Imaging and Superresolution Optical Microscopy
BP 13.10: Poster
Monday, March 16, 2015, 17:30–19:30, Poster A
GPU-based statistical multi-resolution estimators for image reconstruction — •Jan Lebert1, Johannes Hagemann2, and Stephan Kramer3 — 1G. A. U. Göttingen, Fakultät f. Physik, Friedrich-Hund-Platz 1, 37077 Göttingen — 2Institut f. Röntgenphysik, G. A. U. Göttingen, Friedrich-Hund-Platz 1, 37077 Göttingen — 3Max-Planck-Institut f. biophysikalische Chemie, Am Faßberg 11, 37077 Göttingen
We present two implementations of Dykstra’s projection algorithm on NVIDIA’s compute unified device architecture (CUDA). Dykstra’s algorithm is the central step in statistical multi-resolution (SMR) methods (Frick, Marnitz, and Munk, 2012 and 2013) which are a recent development for the deconvolution of noisy images. Unlike other methods its primary parameter is the confidence level with which the reconstruction is considered as valid. Compared with a CPU our CUDA implementation of the standard Dykstra algorithm (SDA) is one order of magnitude faster. For a further speedup we have developed a new variant, which we call incomplete Dykstra’s algorithm (ICD). Implemented in CUDA it yields an additional speedup of one order of magnitude over the CUDA version of SDA. As sample application we discuss preprocessing super-resolution optical fluctuation imaging (SOFI) methods (Dertinger et al., 2009) by ICD. Our results show that a careful parallelization of Dykstra’s algorithm enables its use in large-scale statistical multi-resolution analysis.