Optimizing image processing

[Vaughan]

Hi everyone

I am doing an undergraduate project for image recognition, and I've
implemented my own training system using a variant of Adaboost,
Cascades and Haar Features.

My problem is that I'm experiencing a bottleneck when computing the
weighted error of each weak classifer, since each classifier (of which
there are 200,000+) must be evaluated on each image (of which there
are at least 3000).

I know this is more of a platform specific question, but I was hoping
that someone possibly knows a technique which is used to do this (i.e.
fast image iteration) and is generic enough to be applied to any
platform (such as C#,C++ etc). For example, one may choose to tile all
images together to create a single large image (though this would not
be practicle for my case).

Any insight would be appreciated.

[Chris Silva]

I am a layman at this process, but have been following the group with
interest for over a year. My first exposure with user based clients was
with a company called Riya that was amazing and said they performed 3D
facial recognition and more likely than not found the correct faces and
eerily found many familial and age faces - even in background images. Like
I said, I am not the hardcore scientist here, but interested in the
applications of facial recognition and how it is, and will be applied. Now
Picasa (Google.com) has facial recognition like Iphoto (although I believe
it is better.

My question - has anyone here worked with the Google Picasa solution and do
you know what libraries/processes they are using.

Thank you,
Chris

[yongkang wong]

Hi Chris,

The detail of the algorithm behind Picasa is unknown but I am pretty
sure is the improved algorithm that power IPhoto.

Google acquired a company called Neven Vision 3 years ago, which is
the just after the company licence the algorithm to Apple.

http://www.searchenginejournal.com/google-neven-vision-image-recognition/3728/


Cheers,
Yongkang