Logo detection in screenshot

asked 2018-05-15 06:28:16 -0600

kabeersvohra gravatar image

updated 2018-05-15 08:15:05 -0600


I am trying to detect whether or not a logo exists in a screenshot, outputting a boolean variable. Template matching is not a suitable solution for me since the logos are not exactly the same (obtained from different sources with different resolutions etc). I thought I could use feature matching however this has over matched each logo on each screenshot, providing the same number of matches for a logo on the page vs a logo not on the page. I also tried matching with FLANN but got 0 matches. Results are as follows:

AT&T logo in AT&T page (basic feature matching) C:\fakepath\image0.png

PayPal logo in AT&T page (basic feature matching) C:\fakepath\image1.png

AT&T logo in AT&T page (FLANN feature matching) C:\fakepath\image240.png

AT&T logo in AT&T page (object detection) C:\fakepath\image3.png

Any ideas of what I can try next?


edit retag flag offensive close merge delete


"I also tried matching with FLANN but got 0 matches" -- wrong tool. forget that.

" what I can try next?" -- read about object detection.

berak gravatar imageberak ( 2018-05-15 06:54:31 -0600 )edit

Using the tutorial here: https://thecodacus.com/object-recogni... which also uses feature matching for object detection I get no real results (screenshot added to question), could you suggest something else? Many of the other tutorials are for detecting objects in the image (for example humans).

kabeersvohra gravatar imagekabeersvohra ( 2018-05-15 08:00:24 -0600 )edit

^^ shit tutorial.

berak gravatar imageberak ( 2018-05-15 08:01:39 -0600 )edit

And if I use the good match threshold that they used in the code attached, I get the error: 'Not Enough match found- 1/30'

kabeersvohra gravatar imagekabeersvohra ( 2018-05-15 08:04:36 -0600 )edit

please forget the idea. it's wrong.

(and maybe read books & papers, not blogposts)

berak gravatar imageberak ( 2018-05-15 08:06:31 -0600 )edit

however the official object detection tutorial here http://opencv-python-tutroals.readthe... only shows how to detect things like faces and eyes in images rather than specific logos in screenshots, is there any way of modifying this to do what I want it to do?

kabeersvohra gravatar imagekabeersvohra ( 2018-05-15 08:08:50 -0600 )edit

no, that's not the official documentation, which you will find here

berak gravatar imageberak ( 2018-05-15 08:12:51 -0600 )edit

sorry yes, however the image detection requires to match against a HAAR model, do you think it is feasible to train the model on a series of logos from AT&T and then test the model against the page?

kabeersvohra gravatar imagekabeersvohra ( 2018-05-15 08:19:34 -0600 )edit