One to many comaprision using OpenBR

[Rangarajan kgr]

I am currently workin on a project where the user will compare his face images ( it will be a single Image/Multiple Images ) with the Celebrity images and get the closest match of a celebrity face comapred to his face.

I have gone through the OpenBR document and looks like it is possible with OpenBR. 

1) Can some one help me with the steps to implement the above to proceed with OpenBR? 

2) Also the input images will be like profile pictures of face-book. So how to maintain good acuracy with OpenBR? Do I need to apply any pre-processing steps to the input images and pass to OpenBr? 

Thanks,

Rangarajan

[Brian Williams]

Take a look at this section of the docs for how to do thus from the command line or from your own program. 

http://openbiometrics.org/doxygen/latest/group__cpp__plugin__sdk.html#examples

For more ideas, here are a few sites which do variations on match your face with a celebrity using face recognition, most of them also have apps you can try out for ios and android. 

http://www.picitup.com/NewMatch/

http://celebrity.myheritage.com/celebrity-look-alikes

http://face.naughtyamerica.com

http://celebrity.picadilo.com

http://applemagazine.com/skakash-review-shazam-for-celebrities/6328

http://www.jailbase.com/mobile-app/

Some of these are using openbr for face recognition. 

--
You received this message because you are subscribed to the Google Groups "Open Source Biometric Recognition" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openbr-dev+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Rangarajan kgr]

Hi Brian,

Thanks for the reply. I started implementing the celebrity match. I would like to get few things clarified. 

1) I am planning to take the training images of each celebrity as 5-N images. I dont want to create a gallery file for each celebrity. Instead I would like to 

     create a ".gal" for the entire dataset, by this I can save the time. But I need to take the average score for each celebrity. How can I get that? Or any idea on how      to get the closest match?

2) Most of the celebrity images are of profile pictures and how does OpenBR works for it and how to increase the accuracy? What are the pre-processing steps I need to do before passing the image to OpenBR? 

3) I have read in your paper that the face should be of resolution 128*128? Do I need to check the images before passing to OpenBR whether the face is 128*128 resolution?

4) How to handle pose adjustments to the face using OpenBR, before training/creating a ".gal" file?

Thanks,

Rangarajan