Diploma work: Monthly report 6/97

Review June 1997

I did the following things during the last month:

• I restricted the application of Otsu to a small window around the previous ink position to get a better result. This sometimes still results in a threshold which is too high. To correct this I implemented at the moment an upper limit for the threshold, but in the future I plan to inspect the output parameters of Otsu to get more informations about the look of the histogram.
• Preparation of slides and recognition results for my talk.
• I hold my talk in occasion of the "FKI Seminar".
• I improved the averaging of the black points after the delta image generation. To get a better center in the middle of most of the pixels I use a iterative process. In the first step I calculate a "normal" center where every pixel has the same weight. In the next steps the pixels have weights reverse proportional to the distance from the previous center. With this improvement the detected line strokes look already much better. The letter "e" of the word "Hello" can now be recognized.
• To improve the preprocessing of the images I tried the proposal of Mr. Jiang. In a first step I capture now the greyscale image of the written word "Hello world!", calculate the threshold using Otsu and convert the image to a binary image. This binary image I "thick" about 3 to 7 times and use all the blak pixels as a mask. Only pixels that are black in this mask are treated in the frame sof the image sequence. All other pixels are turned white and ignored. With this approach I can get rid of a lot of disturbing pixels and concentrate on pixels near the written text. The letter "o" of the word "Hello" looks now rather well. Because the time-resolution of the input frames is to high it looks still strange.
• I read in the Diploma work of R. Ammann about the interface to the ETHZ recognition system.

Preview July 1997

My goals for July 1997 are:

• Capture several image sequences of one writer. The written text has to be only a single line, but the resolution over time should be better. E.g. a letter "o" should consist of at least 10 frames to get a nice detection that looks like a round "o". If there’s enough time in July I plan to capture sequences of other writers, too.
• Analyse the output parameters of Otsu to check the reliability of the threshold.
• Test the iterative calculation of the center on several other image sequences. Improve it.
• Test the "mask" on several other image sequences. Improve it.
• Analyse the influence of the ETHZ input interface to my detection process and my output interface.
• Integrate the selection of the area for the mask into my GUI to make testings easier and faster.
• Think about the automatic getting of the mask after the writer completed a line of text.
• 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