Title: Digital Image Processing Algorithms for Cell Centroid and Boundary Detection in Two Dimensions and C
1Digital Image Processing Algorithms for Cell
Centroid and Boundary Detection in Two Dimensions
and Cellular Network Mapping in the Time
DomainStephen Kanji Chen
- Hsinchu, Taiwan
- Tuesday, August 15 2006
2Current progress and goals
Progress on gray-space segmentation was slower
than expected, but debugging was finished today,
and the algorithm is finally running as it is
supposed to. Plans for this final week
- When using the gray-space method on real-life
cellular images to single out nerve cell
centroids, the algorithm often misses a few
pixels here and there from the cell processes,
and sometimes, when identifying the background,
the region bleeds into areas that are actual
cell centroids. There are two simple methods to
counter this - Allow the option for a two-pass filter. The
algorithm outputs a false-color meta-image that
color codes different regions the background,
cell bodies, and cell boundaries. Running the
algorithm again on the meta-image will eliminate
nearly 100 of these pixelation errors. - Add a few lines of code to allow for color bias
during the grayscale conversion process. Since
cells are often dyed a particular color, the
conversion needs to optimize for that particular
color so that no contrast information is lost
during the conversion process. Some of the
issues with region-bleeding is due to poor
contrast. - Test the current image processing algorithms we
have created (boundary detection, cell centroid
detection, image histogram thresholding,
gray-space segmentation) on actual images and
motion pictures from Dr. Silvas lab in San Diego.
3Goals (continued)
- We have decided that due to time constraints and
practical reasons, it is not worth implementing
cell network detection in static images.
Analysis of high-quality movie clips is much more
efficient and accurate. In this respect, we will
begin working on a time-domain analysis of images
next week. - Chances are that a week will only give us enough
time to implement a simple algorithm that has the
basic functionality of a demonstration, but as we
continue to pursue this line of research during
the upcoming academic year, focusing on
developing an accurate and robust cell network
mapping algorithm will be the primary objective.
4Photos of the week
- T minus 48 seconds
- I have yet to understand the reason why many
intersections in the city have these countdown
timers. - If things like these were installed in the US for
the drivers to see, I bet there would be a lot
more incidents of street racing!
5Photos of the week
The 10 meter sprint! Here is another picture
that I took in Hokkaido, Japan of something
similar. The chevrons on the right tick off one
by one. I noticed that conventions for audio
signals for the blind are the same as in the US
honking means clear to cross North/South,
chirping means clear to cross East/West.