AutoHotkey Community

The Font Image Recognition Routine

This routine creates a vector shape sequence file (using -i switch) out of an 100x100 pixels 24 bit colour depth black and white image representing a character true type image (font) or character freehand drawing . Then using the -c switch the two index files derived from two different images are compared and graphics match result is displayed .
For the indexing type:
E.g.: C:\cr13 -i "Drawing1.bmp" "Drawing1_Index.txt"
For the comparison of two different index files type:
E.g.: C:\cr13 -c "Drawing1_Index.txt" "Drawing2_Index.txt"
At present the routine builds shape vectors on black/white bitmaps, it does not support different resolution nor colors/color depth.

But how does it work?

(1) indexing, creates vector txt file (that might be the meta character file) out of the bmp image file in the following manner:
- inverts the b/w file matrix (the way human eye sees it),
- searches for quadrants (10x10 pixels sized) with 40/60% b/w ratio, thus finding character image edges (up to 8 pairs in the same row),
- creates vectors out of each quadrant,
- shifts quadrants by (only) few pixels UP since bmp edges do not always REALLY represent character ID curves, repeating vector creation ...
and
(2) comparison of two vector files:
- shifts back all X-axis values subtracting them by absolute minX value,
- computes curve angles out of each quadrant values,
- computes resultant angles out of quadrant pairs building most real character curves,
- compares the two vector files angle pairs,
- computes match statistics .

This development is aimed for PDA users using easier ways for text input.
To Maria Dzinleska


[Home]