Browse all publications by topic
Browse all publications by year
- F. Model, M. Scholz, and
K. Obermayer. A Hybrid Approach to 3D Blind-Deconvolution of
High-Resolution Confocal Microscope Images.
.
In Proceedings of the 26th Göttingen Neurobiology Conference,
page 780, 1998.
Geometric 3D reconstructions of single filled nerve cells from
confocal image stacks represent an important step towards the automatic
generation of realistic compartment models, which is our long-term goal.
Since the most interesting dendritic and axonal terminal zones are close in
size to the theoretical attainable resolution of the microscope, we find it
therefore reasonable to sacrifice a small amount of speed for maximum
accuracy in the deblurring of these confocal images. We analyze different
deconvolution methods with respect to their convergence speed, accuracy and
robustness. From this analysis we derive a hybrid method which estimates a
reasonably good impulse response (PSF) for the confocal microscope by real
blind 3D deconvolution on a relatively small sample volume and uses the so
found PSF on the whole dataset in a fast non-blind method. bf
Methods & Results:We show results comparing different
aproaches to the deconvolution problem, including a Wiener-type filtering
method, the CLEAN algorithm [1] and both blind and non-blind Richardson-Lucy
algorithms [2]. In the case of non-blind algorithms the PSF is determined
experimentally with small fluorescent beads, which are embedded in the
tissue. Whereas the Richardson-Lucy algorithm is designed to perform best in
regimes, where photon-shot-noise is dominant (under bright light
illumination), the CLEAN algorithm seems to perform better in regimes, where
sensor-noise due to heating is the limiting factor (whith fading dyes and/or
low light intensities). In terms of accuracy the blind methods perform best,
since they do not suffer from poor estimates of the PSF. On the other hand,
blind deconvolution on large datasets is almost not feasable, due to
computional expense. We therefore propose an algorithm which combines the
best of two worlds, by using blind deconvolution on small sample volumes to
get good estimates for the PSF and then deblur the whole datavolume with a
computationally less expensive derivative of an non-blind algorithm. [1]
Schwarz, U.J. amp;quot;Mathematical-statistical description of the
iterative beam removing technique (method CLEAN)amp;quot; Astron.
Astrophys. 65, 345-356 (1978), [2] Richardson, W.H. amp;quot;Bayesian-based
iterative method of image restorationamp;quot;, J. Opt. Soc. Am, 62
Supported by BMBF grant Nr.: 0310962.
|
