Ask Your Question

Improve Brand Detection Quality

asked 2014-09-25 16:35:16 -0600

johncordeiro gravatar image

updated 2014-09-27 07:55:27 -0600

I am trying to detect brands in pictures but I'm having some problems.

I did the cascade training with 500 positives samples and 1440 negatives samples, but I'm getting a lot of bad results. The cascade classifier detects the brand and further a lot of things that isn't the brand.


I tried to use SURF features to eliminate these bad results and look that it isn't work:

image description

The same result occured with a image that is not my brand:

image description

Look the detections:

image description

My question is: How do I eliminate the bad results and just stay with the brand that I am looking for ? What I'm doing wrong ?

edit retag flag offensive close merge delete


Are you looking for a brand logo or something else? Can you explain a bit more, like how big is the object you are trying to recognize in absolute terms as well as in relation to other objects in the image.

unxnut gravatar imageunxnut ( 2014-09-25 17:55:29 -0600 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2014-09-26 00:23:55 -0600

updated 2014-09-26 00:29:31 -0600

Hi! You can go with opencv feature detection algorithms such as SIFT.

Have a look at the feature detection module of opencv. Play around with that & come up with some results.

Also have a look at this thread.

edit flag offensive delete link more


I followed your advice and I edited my question. My results is getting better with SURF than SIFT. It didn't solve the problem yet.

johncordeiro gravatar imagejohncordeiro ( 2014-09-27 07:59:57 -0600 )edit

Implementing SURF also involves filtering the matches, i.e., finding good matches. Checkout this link,

sumitsrv gravatar imagesumitsrv ( 2014-09-29 09:36:42 -0600 )edit

I'm doing exactly in this way, and the results are these in question.

johncordeiro gravatar imagejohncordeiro ( 2014-09-29 11:57:52 -0600 )edit

Question Tools


Asked: 2014-09-25 16:35:16 -0600

Seen: 642 times

Last updated: Sep 27 '14