Comments on: Face detection now works with Mono

By: Ali Servet Donmez

Ali Servet Donmez — Sat, 24 Oct 2009 12:31:22 +0000

I’ll do my best, thanks for your interest and I’m looking forward to hear from you on f-spot-list. Regards.

By: etienne

etienne — Fri, 23 Oct 2009 22:34:08 +0000

Thank you for your answer. I understand more what “thesis” means in your context ;-)

Implementing the solutions presented in the paper I found would probably take a few weeks, full-time. And I _really_ can’t be full time. And I used only once opencv, for a very small demo !

The first thing that I can say is, please integrate the opencv calls nicely, so that this part could be further optimized with newer face detection algorithms. And extended by face recognition.

I will register to the f-spot list. I think it really can take place there.

By: Ali Servet Donmez

Ali Servet Donmez — Fri, 23 Oct 2009 17:15:45 +0000

As being a thesis student, I could probabily use those facilities for free. Honestly, I’m not working on this full-time, nor I did time to make any research on the subject. Let’s say, the first target was to build and run a sample OpenCV code in C# and this is what most importantly achieved here.

The next target would be, getting my hands dirty on F-Spot code base, and implement region tagging in a end-user friendly way.

If you could hand me over a ready to run program in order to compare AdaBoost’s results with my simple test images, I’d appreciate that. Nothing is written on stone in this project, so if AdaBoost’s results are clearly better than mine and it’s relatively easy to implement, I would use that one instead.

This would be an undergrad thesis project and my main focus is to come up with something that works in very a limited time.

I believe you could use F-Spot-list, I’m subscribed to that one, but there are no discussion on this topic at the moment. Further more if they see that such a discussion won’t fit in that ML, we could definitely find some other place to do that.

By: Ali Servet Donmez

Ali Servet Donmez — Fri, 23 Oct 2009 16:54:47 +0000

Thanks!

By: etienne

etienne — Fri, 23 Oct 2009 15:00:14 +0000

I am a ph.d student (not in this field !) but have done some pattern recognition courses.
I have access to the IEEE xplore through my university, do you have it ? I searched yesterday some papers about face detection, and found easily one recent paper, serious, quiet complete I think. Maybe an interesting solution to implement with opencv ?

Face detection in color images using AdaBoost algorithm based on skin color information
Yan-Wen Wu, Xue-Yi Ai (2008)

Most of today’s pictures are colored, so I think this is interesting solution (when looking at their results, it seems interestingly better at face detection than the one in opencv – looking at your results).

What do you think ? I suppose you will focus now on region tagging. Are you full-time on this project ? (I saw an input in your website saying it will be your thesis – master or Ph.D or something else ?)

Is it possible to follow/talk about your developments on a mailing list ? I will register to it to follow your work. I am definitely interested in such feature, and I think I can help a bit. And of course encourage ;-) !

By: Emre Sevinç

Emre Sevinç — Fri, 23 Oct 2009 11:43:07 +0000

Cool project. I’m expecting much more from you ;-) Such as face-recognition all the way down!

By: Ali Servet Donmez

Ali Servet Donmez — Fri, 23 Oct 2009 10:35:24 +0000

Hey, thanks for your comment. My first target was to learn how to PInvoke OpenCV library calls in Mono C#, so the code above is just a re-write of the example face detection implementation suggested by OpenCV book, there’s not much extra work has been done to improve the results.

Face recognition definitely is a must, but not a priority at this point. First of all, I’ll do my best to get basic face detection and region tagging implemented for F-Spot.

Any help is appreciated.

By: etienne

etienne — Fri, 23 Oct 2009 02:26:05 +0000

Interesting.
False positive seems quiet low, but the algorithm seems very sensitive to face orientation. Faces in front of the camera, with low tilt seems very good, otherwise… much harder for the algorithm to properly detect faces.

Which algorithm do you use ? openCV library ?

Do you plan to get involved in face recognition ?
And more importantly, what are your plans for face recognition ?

Thank you.