Diploma work: Monthly report 5/97

Review May 1997

During May 1997 I worked mostly on improvements of the preprocessing and the generation of delta-images. As already announced in the last report at the moment the biggest problem is the guessing of a good threshold for the black and white conversion. This threshold should be adapted for every single image.

Mr. Ha recommended the use of the algorithm of Otsu on the Histogram. The description of this algorithm was very promising. But it didn’t succeed on my delta-images. Because my delta-images consist of mostly white pixels and only some dark pixels the threshold of Otsu is in the middle of the bright pixel values. A threshold like this results in a useless image.

All the Histograms of my delta-images have about 98% of the pixel values in the range between 200 and 255. The remaining about 2% of the pixels are distributed in the grey and dark level ranges of the Histogram. Then my idea was to compute a threshold in the way that the darkest 10 pixels become black and the remaining pixels turn white. If there is ink that has to be detected on the image this method works fine. But if there’s no ink that has to be detected the darkest 10 pixels (which are indeed a rather bright grey) are converted to black and are misdetected as ink.

Because the above two methods didn’t help I started to focus the detection of ink in a window. This window is 40x40 pixels big and placed around the last detected position of ink. With this method a lot of disturbing pixels, mostly from the pen or the hand are out of the window and not detected. To get enough dark pixels out of the grey ink traces the threshold for the black and white conversion is in the upper half of the Histogram.

As a last improvement I record every ink trace that was used for the detection in a special image. After the generation of a delta-image all the pixels that were already detected are deleted ot ouf the delta image. This is very helpful if pixels that were already detected are reapearing. This happens when the pen tip covers a bit of a ink trace in one image and is moved away in the other. The normal delta-image generation detects this a a new painted ink trace, but this correction prevents from redetecting such ink traces.

Preview June 1997

My goals for June 1997 are:

• Still improve the generation of the delta-images. I don’t know how at the moment. But I think that I have to try to improve the inputs to the delta-image generation and do some corrections on the output.
• The approach with the window has one main problem: When the new appearing ink trace is outside of the window no more ink traces are detected. To prevent this I have to do a backtrack if I don’t get new ink traces for some time. An other problem is the detection of the first ink trace. For this I first have a window that covers the whole image. If I detect wrong ink traces in the first image the whole process fails. Perhaps I have to get some user input in the start image.
• Read about projective geometry to be able to reproject the captured image into a normalized view of the paper. With this I should be able to detect the position of the paper and normalize into the point of view of a scanner or a graphics tablet.

zurück zur Uebersicht der Diplomarbeit

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