Compare images with symbols (hand written/drawn)

[Phaser]

Hi Guys

This I think is going to be a tough one, I have managed to crop my images on all 4 sides to the edges of the symbols I am working with. Our stock sheets, that we already scan and read succesfully with Autoit, and of course the thanks to Melba & many others in the past, well stock sheets change all the time and so far I have not hit a stumbling block until now.

Management added a small box to the forms for our hired help to manually write either a cross, line sloping right or left or a circle, we still scan the forms and process them with autoit but I need to KNOW what the hired help did in that box.

I have attached 3 pairs of examples for you to see what I am dealing with, and how different they can be but when checked I need them to be matched succesfully each time.

I have looked at the threads dealing with image matching but they mainly work on pixel differences, so they throw high percentages of difference most of the time, is there a better way? I suppose OCR has a few tricks to read different font faces does it work on the shape of a letter or does it try to work out the font then compare with its DB, do I need to create templates for each symbol?

Any ideas would be most welcome, thanks for looking.

1a.bmp

1b.bmp

2a.bmp

2b.bmp

3a.bmp

3b.bmp

[alienclone]

perhaps instead of looking at the small box as one single image, break it into 4 images, one for each corner.

if corner top right and bottom left have pixel changes, then it would be example 3.

if corner top left and bottom right have pixel changes, then it would be example 2.

if none of the four corners have pixel changes, then it would be example 1.

if all of the four corners have pixel changes, then it would be a circle. (supposing the circle you mentioned was to be drawn inside the square, since you did not give and example of that.)

[Phaser]

Thanks for the reply, I did think of that but they don't always hit the box fully, sometimes half in it and half out. The box is easy for me to remove from the scan as its created by a computer, but I digress, the human element makes it tough.