Logo Recognition with logo database

[Adnan Rashid]

Hi,

I have never used Accord.Net before, and have only gone through the examples and am pretty excited by it. Currently I am using EmguCV for my CV tasks, as I am pretty comfortable with C#. Unfortunately I have run into an issue and can't find the solution or even verify the correct approach.

I would like to perform logo recognition across a database of logo images. I have been using the EmguCV SURF example and found good results for one vs one image recognition. Unfortunately in my case, my logo database can go up to a few hundred and a one vs hundred scenario is not scaling well. I read a few research papers, and found that it would be better to build a bag of visual words (vocabulary tree). Unfortunately I couldn't find any code reference for this, to test it out.

I would like to know if this is possible (or easier) using Accord.Net and if there is any article or code reference to get me started.

Thanks in advance.

[César]

Hi Adnan!

Perhaps you could take a look at the Image Classification Sample Application. It uses a Bag-of-words generated from SURF features, plus KSVMs to perform the classification. Please keep in mind that all of those points are switchable (i.e. you can exchange SURF with FREAK, for example, or other detectors) and the KSVMs with any other classifier of your choice.

If you have downloaded the framework, the application should already have been included on the installation path; otherwise you can also download it directly or browse the sources online.

Hope it helps!

Best regards,

Cesar

[damola....@gmail.com]

Hello Cesar,

I was wondering if you had any advice on using the Bag-Of-Visual words compute method with 5,000 training images? Is there an efficient way to do this? Is it even possible with the current implementation? How would one manage memory considering the implementation takes a Bitmap array?

Thanks,

Damola