Diploma work: Monthly report 12/97
Review December 1997
I did the following things during the last month:
- I captured 7 different image sequences written by my
girl-friend and me. Now I have a new test set with 15
lines of text with about 50 English words. Unfortunately
all the compressed MPEG sequences take 1GB of diskspace
and decompressed they take about 4GB ... !
- I implemented a new detection method based on the
distances of pixels. First I calculate for every pixel
the sum of the distances to the other pixels. The result
is the pixel with the smallest sum. The results of this
method are good and there are no new pixels generated
like with a simple center.
- To decide whether a result of Otsu is reasonable or not I
added a new criteria. Until now I decided this based on
the variance of the histogram, but this approach showed
problems with some delta images. The new criteria is
based on the bandwidth of the histogram. Because every
delta image has a lot of bright pixels I can limit the
usage of Otsu to histogram with darker parts. This
criteria is as good as the variance criteria but
it’s more intuitive for the user because
there’s a direct correspondence to the values and
scales in the histogram.
- I started the analysis of the new image sequences, but I
don’t have concrete results yet. Some sequences show
problems with the contrast of the ink traces. A part of
those problems could already be solved, the others are in
progress.
- Dr. Schenkel from the ETHZ told me that they have a
Demo-CD of their system. This demo version is able to
process direct input from the tablet or generated
databases. It could be possible that we can generate the
database offline from their system and then either
process it locally or remotely in Zurich. I will make a
date with him to fix the procedure and test and learn the
system.
Preview January 1998
My goals for January 1998 are:
- Prepare and give the talk for the seminary on the 20th
January 1998.
- I want to get good results on most of the new 14 (2 * 7)
images sequences and of course on the most recent old
ones. The several parameters for the sequences have to be
unified as much as possible.
- Meeting with Dr. Schenkel from the ETHZ to talk about the
procedure for the test with the online-recogniser. On
this occasion I’d like to test some of my sequences
at the ETHZ to be ready to eventually use it myself for
further test procedures.
- Preparation of my output for the ETHZ system.
- If the above shows satisfying results I’d like to
check a slightly new approach. I’d like to analyse
the greylevels of a single pixel or a set of pixels over
all the frames. In the optimal the greylevels should be
white until the ink covers this pixel and then stay
black. If this pixel is covered by the pen in the
meantime there will be a short disturbance of the graph.
zurück zur Uebersicht der
Diplomarbeit
Last modified 23.12.1999 00:18:51 by Thomas von Siebenthal