0

I am using OpenCV on iOS to detect a rectangular label to assist users in snapping a photo of that label. I have an overlay that presents once the matches threshold is met.

My question is, does that patch image used have to be exact? The labels I am detecting have text on them that vary from label to label. All the same font but different characters. Is it possible to train OpenCV with a patch images color and/or size/dimensions? Or is there perhaps another way around this issue?

Here is a close example to the labels Im scanning, EXCEPT THERE ARE NO ICONS AND ALL ONE FONT TYPE.enter image description here

Here is the tutorial I am following, which is achieved with an image of a target. http://www.raywenderlich.com/59999/make-augmented-reality-target-shooter-game-opencv-part-3

SprayKid982
  • 333
  • 3
  • 16
  • So essentially you're detecting coloured rectangles? – a-Jays Apr 24 '15 at 05:55
  • @a-Jays Correct, I am aiming for something similar to the card.io implementation but I couldn't not find any decent direction on how to implement edge/border detection for the label. I am much heavier on ObjC than C++. – SprayKid982 Apr 24 '15 at 14:21
  • You might want to consider some shape detector approaches. Template matching might work, but your templates have to be pretty exact. A colour histogram based method is also something that could work. – a-Jays Apr 24 '15 at 15:43
  • Thanks, is there anywhere you can guide me that might be able to help? The goal is to have a capture rect and then once filled with subImage (i.e. close enough to capture, and perform OCR on), then capture. All the shape detection tutorials I found simply detected them but had no matching context to a master/capture frame. – SprayKid982 Apr 24 '15 at 17:15
  • Do you know the color beforehand? – Rosa Gronchi Dec 10 '15 at 11:05

0 Answers0