Fingerprint matching algorithm using shape context and orientation descriptors

Fingerprint matching code using a hybrid descriptor. EER < 1% (approx. 0.75%) on FVC2002 Db1_A.

Description

This is a fingerprint matching code based on the fingerprint matching algorithm found in the upcoming book chapter (published in July 2011):

Book title: State of the art in Biometrics (ISBN 978-953-307-489-4)
Chapter title: Fingerprint Matching using A Hybrid Shape and Orientation Descriptor
Authors: Joshua Abraham, Paul Kwan and Junbin Gao

Located at:

http://www.intechopen.com/articles/show/title/fingerprint-matching-using-a-hybrid-shape-and-orientation-descriptor

Note: LATEST version of the book chapter describing some minor simplifications to the implemented algorithm is now attached in the sc_minutia.zip file (with identical performance recorded, EER=0.75% for FVC2002 DB1_A). Please still reference the original book chapter found at the link above.

Performs reasonably well on FVC2002 database 1 set A. This algorithm uses a hybrid descriptor based on the shape context and Tico's orientation descriptor. However, the shape context descriptor has been modified to include Gaussian bin weighting and other modifications. The Tico orientation descriptor has also been modified to weigh orientation point coverage.

• Required Products: Image Processing Toolbox
• MATLAB release: MATLAB 7.4 (R2007a)

[Bonnie Lee]

This is very good software with low EER. However, documentation
needs fixing. The readme needs updated.

I had to add path sc_minutia folder first.
After templates created, I can run
BASE_IMG=''; main

This is long however. But I found that Ctrl-C to interrupt
process can be continued again with just 'main'.

also, when stopped, the RES_G and RES_B arrays can be used
to plot EER graphs with

[ig,ib]=calc_EER(1./RES_G, 1./ RES_B);
plot(ig);
hold;
plot(ib, 'red');
axis([0 200 0 100]);

The EER is quite good. I have a question, what are the SC_G
and SC_B vectors for?

[Joshua Abraham]

SC_G and SC_B are additional shape context similarity distance.

They can be added to the the RES_G, RES_B scores, for instance
with something like:

 roc( RES_B + 0.1*SC_B, RES_G + 0.1*SC_G)

Thanks

Joshua

[Joshua Abraham]

Just one more thing to mention, this is proof of concept code.

The code will be further documented and optimised. At the moment, it runs slow (sorry about this) and documentation is
sparse. Please, if you have any questions about the code, email
me.

[Joshua Abraham]

Sorry, mistake...I should have written

 roc( RES_B - 0.1*SC_B, RES_G - 0.1*SC_G)

before, since we are dealing with distances not similarity scores.

[Paul Smyth]

Hard to find good performing code on the net. Thanks for this.

[JuCheng Yang]

it is a nice code, one note, i meet a bug in the do_match.m;

??? Error using ==> load
Unknown text on line number 1 of ASCII file 'F:\sc_minutia\'101_1.o'

Error in ==> f15 at 1
load('F:\sc_minutia\'101_1.o');

to solve it, change all the "load" into "importdata" in the file do_match.m.

[Joshua Abraham]

Thank you JuCheng for your bug notification. I will make the changes you suggest in the next update.

[Le Hai]

Hi Joshua,
Could you give your idea about the combination between shape context and orientation descriptions in your code, so we can follow your code easily.
Thank you.

[Joshua Abraham]

Hello Le

The good news is that a book chapter is about to be published on June 21st 2011,
which will be free for everyone to download. Unfortunately, I am not allowed to give detailed comments until this date. However, I will provide a link to the book and attach the book chapter at the publishing date (21st June).

I hope this will help everyone.

[Le Hai]

Hi Joshua,

Is the link to the book chapter available and free for everyone?

Thank you much.

[yagnesh]

 

[Abioye Samson]

Hi Joshua, i am also working on a similar project, Thanks for the help. God bless u.

[Abdul Aziz]

Excellent work!!

[nejc]

Hi! I get an error for folowing line of code: "file_a = substring(char(file_a), 0, findstr(char(file_a), '_')-2);". Matlab returns an error: "??? Undefined function or method 'substring' for input arguments of type 'char'.". There is no substring function in matlab... How did you solve this ?

[nejc]

Ok, stupid question... I solved it :) will be any speed up of the algorithm available in the future ? :) Thank you

[Joshua Abraham]

Hello Nejc,

Sorry for the late reply...I have a much much faster algorithm coming out soon.

The EER for that is 0.96% to 0.67%..

This will be uploaded soon.

Thanks

[nejc]

Joshua this is great! Can't wait for new release!

Thank you!

[Sushobhit Rajan]

means comparing the template is gud..and thanks for ur logic....but now how should i take input through device and check the identification

[Joshua Abraham]

Sushobhit, this code is aimed towards fingerprint images already collected by scanning devices. You may need to write some trigger code that collects an image file from the scanners and then calls this code for feature extraction
with 'extract_finger <image file>'. Just add a line to this script change to the path the images will be saved. Hope this helps

[harjan]

hi, What is the term COMP_IMg meant?

[Joshua Abraham]

Hello Harjan, COMP_IMG is the string reference to the database for the query fingerprint impression.

[zaidan]

let me know it first but it's nice code for beginner

[Sesh]

Joshua,can you tell me the reference as to how in the testfin.m,you could say the median frequency value usage gives better output??I read different papers but could not find it anywhere..please help..