Diploma work: Monthly report 11/97

Review November 1997

I did the following things during the last month:

• I improved the Interpolation routines. I can now interpolate between a defined number of gaps in the middle of two pen downs but the distance has to be smaller than a certain value.
• The analysis of the results showed that the "Simple" Detection method which just calculates the center of black pixels shouldn’t be used. It’s too unstable. The same is true for the "Greyscale" detection method which weights the positions based on the grey level.
• The size of the mask to cut out the interesting parts of the text should be small. 2 seams to be good.
• Interpolation can improve the results a lot. But the parameters have to be adapted from image sequence to image sequence. But it’s very easy to test several settings, because the redrawing can be done on the fly with nearly no delay.
• The settings for the Lookahead and the Limit for the Otsu-Variance are now restricted to a certain interval but I didn’t find yet a "perfect" value.
• I got from Dr. Schenkel of the ETHZ the confirmation that he would help me on one day to use the recognition system at the ETHZ. Actually there’s no person at the ETHZ using this system.
• A PERL script allows me to launch a lot of program instances on different computers to calculate an image sequence with different parameters and detection methods and generates a HTML Web Page.
• Because I limit my calculations to a small bounding box I could speed up the whole processing. With the fastest machine (Pentium II 266MHz, Linux) I got a throughput of about 2.5 frames/s.

Preview December 1997

My goals for December 1997 are:

• Improve the selection and combination of parameters and perhaps the program routines to get good results over all image sequences. The latest sequencies (slow written, good contrast, large letters) are done well, but some worse (either bad contrast or written very fast) show problems in detection.
• Improve the conversion from the point information to the line strokes. At the moment I improve the conversion only with interpolation and the suppressing of too large linestrokes.
• Get some more image sequencies (good quality, other writers, ...) that can be tested.
• If the above showes satisfiying 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 coverd by the pen in the meantime there will be a short disturbance of the graph.
• Preparation of my output for the ETHZ system.

zurück zur Uebersicht der Diplomarbeit

Last modified 23.12.1999 00:18:51 by Thomas von Siebenthal