Resolution of the image for fast data transfer

[visionm...@gmail.com]

I am capturing image at the client side and sending it to the server where OpenBR is installed. 

I need to know at what resolution I can capture the image at client side and send it to OpenBR keeping low-resolution and high data transfer in mind, without compromising the accuracy a bit.

can you define it in terms of (W*H)?

[Charles Otto]

For face recognition, the most relevant measure is the resolution of the face in the image, not the size of the entire image. I think Josh previously recommended a inter-pupil distance of 44 pixels (the image should still contain the entire face)--Josh was that number more like a minimum or an ideal value?

Anyway, what having some specific IPD will mean in terms of the image resolution will depend on what proportion of the image is actually covered by the face. If the face takes up a small fraction of the overall image, you will need to transmit a larger image than you would if the face took up most of the image--unless you can crop the image down to just the face.

I don't know what your capabilities on the client side are, but you might also consider doing feature extraction client side and transmitting the extracted features to the server. That would be the best way to minimize data transfer, but requires some computation on the client.

-Charles Otto

--
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/groups/opt_out.

[Joshua Klontz]

Charles, we register face images to 44 pixels between the eyes, so I would consider that value a _minimum_. I can't think of a situation where having more pixels would hurt.

-Josh

[visionm...@gmail.com]

How to find the interpupil distance between two pupils. Can I get a short description?

[Charles Otto]

It's just the distance in pixels between the centers of the eyes. If your imaging conditions are pretty stable you can just take a single image, manually find the centers of the eyes and get the distance from that.

-Charles Otto

[Brian Williams]

You have two optimization criteria: 

1. the size of the image transmitted, the "best" solution here is to send over only the cropped face image with an IOD>44
   
2. the method used to compress the transmitted image, best practice is 0.3 bpp (i.e., compression ratio of 26:1) using a lossy codec like JPEG - if you want to do better, you can move to JPEG2K and incorporate a custom face mask during the compression.

Quite a bit of literature exists on this problem (for both still and video) - as it is one of those easy to define studies with clear metrics.  Here are some older ones, also NIST has some newer ones.

• Delac, K.; Grgic, M. & Grgic, S. (2005). Effects of JPEG and JPEG2000 Compression on Face Recognition, Lecture Notes in Computer Science, Pattern Recognition and Image Analysis, Vol. 3687, 2005, pp. 136-145
  
• McGarry, D.P.; Arndt, C.M.; McCabe, S.A. & D'Amato, D.P. (2004). Effects of compression and individual variability on face recognition performance, Proc. of SPIE, Vol. 5404, 2004, pp. 362-372

and one on video from a member of the core openbr team:

Assessment of H. 264 video compression on automated face recognition performance in surveillance and mobile video scenarios
B Klare, M Burge - SPIE Defense, Security, and Sensing, 2010