Real-life receipt parsing issues

[LecrisUT]

Would be helpful to keep track of some common real-life receipt issues. Although It might be better as a Github issues so each issue can have its own thread to discuss on.

• [x] Edge detection is sensitive to lighting and background contrast (e.g. pure white receipt on grey or coloured desk is not detected)
• [x] Cropping the receipt cuts off part of it
• [ ] Detection on non-latin alphabet? (to be tested)
• [ ] Receipts can be badly crumbled? (to be tested)
• [ ] Receipts will generally be curved? (to be tested)
• [ ] Some receipts are coloured: e.g. credit card or train ticket ones. Since those are more niche usecases and have no item information it can be ignored for now.
• [ ] East Asian receipts are very packed both in design and character design so it can be sensitive to focus, blur and grey-scaling.

[monolidth]

Great ideas and I agree with all of them and I try my best :)

[monolidth]

Update, I tried different methods and parameters e.g. I tried to use the otsu method to calculate the best threshold but with no success, maybe this is not so trivial :)