Thanks

I've been using it more and more and the accuracy varies qutie a bit depending on the quality of the photo (obviously). I've had it get a board 100% correct but also had it where you have to change around 7 or 8 of the clues.
I think I'm happy with it though. I may be able to improve it in a later version but I'm going to stick with what I've got for now.
At the moment it does this...
1. Find the board within the picture.
2. Adjust the perspective and crop down so that only the board remains.
3. Split the board into 81 smaller images (i.e. an image per cell).
4. Try to remove anything that isn't the number from the image.
5. Try to recognise the number.
1 and 4 use the same method which is to find the "biggest blob" within the image. In step 1 the biggest blob will be the sudoku board. In step 4 the biggest blob will (hopefully) be the number. Once the biggest blob has been found it then fills in everything else so that only the biggest blob remains.
(I just realised that this may not be fully working. Will have to investigate tonight.