GPU implementation of map-MRF for microscopy imagery segmentation
The Institute of Electronics, Communications and Information Technology, Queen’s University Belfast, Belfast, UK
IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2009. ISBI ’09
@inproceedings{crookes2009gpu,
title={GPU implementation of map-MRF for microscopy imagery segmentation},
author={Crookes, D. and Miller, P. and Gribben, H. and Gillan, C. and McCaughey, D.},
booktitle={Biomedical Imaging: From Nano to Macro, 2009. ISBI’09. IEEE International Symposium on},
pages={526–529},
year={2009},
organization={IEEE}
}
Recent developments in 3D low-light level CCD (L3CCD) image capture have enabled the study of the dynamics of biomedical bodies within cells. This paper firstly presents an improved algorithm for automatic segmentation of such imagery. It allows for the specific nature of noise in L3CCD data. Secondly, the massive volume of data produced by continuous real time 3D scans requires a high performance computation facility for automatic segmentation and tracking. The paper presents details and results of a GPU implementation of a version of the segmentation algorithm, and shows that on an NVIDIA GeForce 8800GTX, coded in CUDA C, the algorithm runs around 550 times faster than the Matlab version of the algorithm running on a PC.
May 25, 2011 by hgpu
