Hi All,
I'm tasked with finding a way to identify shop on printed receipts. It is to check if shop logo / id details from given receipt match any of from the collected ones already. Shops often have logo and some text as part of top section of receipts and it has to match all these details. It is like logo matching.
Methods I've reviewed so far:
- Haar clasifier - seems to be an interesting one but it requires a lot of training and in this project, at the beginning it will star with one or two sample images only rendering training process impossible.
- templateMatch - seems to be the right approach but because template and the tested receipt image will differ slightly (angle, lightining, etc.) it won't be one to one match and if my understanding is correct templateMatch is to be used with sample coming out form tested image.
- SIFT - probably the right approach.
The aim is to achieve:
- new receipt image with logo is being provided
- application checks if any of logo templates from DB matches the one on the receipt.
- if result is positive, it identifies which one.
- once matched, the DB might be updated with new picture of existing logo to have more samples of same logo for better matching next time.
On algorithm level, it will have to be SIFT comparison one by one to all collected logos in DB and select highest ration match.
Any help and or directions will be more than welcome.
Thank you in advance, David