ADR algorithm question
Posted: Sat Apr 22, 2017 7:29 pm
Can you give me a rough idea about how the ADR algorithm works so I can present better images to it for recognition? I'm trying to recognize US tax return forms and I get quite a number of false matches. I've added up to 8 images into a template with still no luck. The documents I'm trying to match are scanned and have some horizontal and vertical offset, skew, and even some compression. Do I need to address the skew and offset prior to calling ADR? What about compression? I'm not clear what is "important" to ADR in ensuring the best chance of getting a match.