Path: news.mathworks.com!not-for-mail
From: <HIDDEN>
Newsgroups: comp.soft-sys.matlab
Subject: Re: New Iris Recognition Results
Date: Mon, 18 May 2009 19:04:02 +0000 (UTC)
Organization: 'SpiruHaret' University, Bucharest
Lines: 31
Message-ID: <gusbf2$65b$1@fred.mathworks.com>
References: <guoduq$ckt$1@fred.mathworks.com> <ff4417e1-0a28-41da-8d35-6c0fa470d048@s16g2000vbp.googlegroups.com> <b50dafbc-09c5-4556-bdfd-c2a947748a6d@n8g2000vbb.googlegroups.com>
Reply-To: <HIDDEN>
NNTP-Posting-Host: webapp-05-blr.mathworks.com
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
X-Trace: fred.mathworks.com 1242673442 6315 172.30.248.35 (18 May 2009 19:04:02 GMT)
X-Complaints-To: news@mathworks.com
NNTP-Posting-Date: Mon, 18 May 2009 19:04:02 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 872427
Xref: news.mathworks.com comp.soft-sys.matlab:540807



Hello ImageAnalyst,

Hamming Distance vs. PSNR:

Let consider PSNR first. It is usualy used as a measure of image distortion. 
Typically, PSNR is used to evaluate the quality of a reconstructed image when compared to the original image. 
More general is the case when one of the images is considered a noisy approximation of the other but the same object is represented in both images.
This assumption is not true for iris images. 
Here the most important difference between two iris images comes from the fact that, even in the most favorable practical scenario, two iris images represent two distinct hypostases of the iris i.e. two distinct objects.  
Comparing two iris images using PSNR doesn't make much sense if accurate motion law IrisFeatures-PupilDilatation is not available. But comparing two iris binary templates using PSNR or Hamming Distance is the same thing. 
There are a lot of papers investigating how image compression  (i.e PSNR) modifies the recognition accuracy.

Hamming Distance vs. cross-correlation:

I still study this subject. I belive that interesting results will arise from this area (at least in terms of testing the robustness of any recognition method).
As an example, the following figure shows some recognition results obtained durring the previous month for Bath University Iris Database when I used a custom modified Hamming-correlation distance:

http://www.limbistraine.com/ro/software/popescu-bodorin/img/bath-university-iris-database--recognition-results--custom-modified-hamming-distance--popescu-bodorin-nicolaie.gif

But, despite the above results, bigger computational effort is a strong reason for avoiding the use of correlation in real-time applications.

Regards,
Bodorin  

______________________
ImageAnalyst wrote:
> Can you expain why Hamming distance is used to compare the "test"
> image to the "template" iris image?  What advantages does that have
> over other image comparison methods such as PSNR, cross correlation,
> or similar kinds of things?