Gpu Implementation of a Poisson Map-mrf Segmentation Algorithm for 3d Tracking of Low Light Level Imagery

The high sensitivity of these sensors provides us with the ability to acquire multidimensional datasets at high temporal sampling rates under low-light level conditions. This allows, for the first time, the study of the dynamics of biomedical bodies such as cells, and their intracellular components. Novel statistical distributions are present within the data due to both the initial low-light levels at which the data is acquired, and the avalanche multiplication to which the data is subjected via the EMCCD. The size of these datasets, coupled with the statistical distributions, present a new challenge within the area of biomedical image processing. Related work in this area has included the segmentation of medical images using an automated volumetric segmentation system [1], and the segmentation of 3D MRI brain images [2]. While these works have adopted a Gaussian MAP-MRF approach, we instead assume low-light level Poisson statistics for the optimisation of the novel L3CCD data; and apply the resulting technique.