DOWNLOAD SITE

 

Instructions

Instructions for downloading a database

 Except check database (see at the end)

 

Download the license agreement (next to the database description), fill it out, sign it, and fax it to +34 928 451 243 (Attn.  Miguel A. Ferrer)

 

Send an email to gpds@gi.ulpgc.es, as follows:

Subject DATABASE download

Body: Your name, e-mail, phone number, organization, postal mail, Database you require, purpose for which you will use the database, time and date at which you sent the fax with the signed license agreement. Along with the email send a pdf file with the signed license agreement.


Send by postal mail the original of your signed license agreement to the following address:

Miguel Ángel Ferrer Ballester

Departamento de Señales y Comunicaciones

Universidad de Las Palmas de Gran Canaria

Campus de Tafira s/n

35017 Las Palmas de Gran Canaria, SPAIN

 

Once the fax and email of the license agreement have been received, you will receive an email with instructions to download the database. After you finish the download, please notify by email that you have successfully completed the transaction.

 

For more information, please contact: gpds@gi.ulpgc.es

 

TOOLBOX
   

gpds HMM

A Hidden Markov Model (HMM) Toolbox within the Matlab environment is available. In this toolbox, the conventional techniques for the continuous and discrete HMM are developed for the training as well as for the test phases. The ability to make different groups of components for the vector pattern is provided. Multi-labeling techniques for the discrete HMM is also provided. The toolbox includes procedures suitable for the classical applications based on the HMM, as pattern recognition, speech recognition and DNA sequence analysis.

 

This toolbox is distributed as binary (dll files) and source code format. For a wide promotion, we ask the users to make a reference to the paper:

Sébastien David, Miguel A. Ferrer, Carlos M. Travieso, Jesús B. Alonso , "gpdsHMM: A hidden Markov model toolbox in the Matlab environment", accepted in Int. Conf on Complex System Intelligence and Modern Technology Application,  Cherbourg, Normandy, France, September 19-22 2004.

 

For any remarks about this toolbox, do not hesitate to contact the authors sending an e-mail to: gpds@gi.ulpgc.es

 

 

Paper Referent    
user's guide    
    download  gpdsHMM  
DATABASE

gpds HAND

GPDS150hand database

 

The database consists of 10 different acquisitions of 150 people by a desk scanner. The 1500 images have been taken from the users’ right hand. The user in our system can place the hand palm freely over the scanning surface; pegs, templates or any other annoying method for the users to capture their hands are not used. The hand contour with landmarks (valleys and tips of the fingers) and the segmented palms are also provided. They have been obtained automatically without supervision as described in:

Miguel A. Ferrer, Aythami Morales, Carlos M. Travieso, Jesús B. Alonso, “Low Cost Multimodal Biometric Identification System based on Hand Geometry, Palm and Finger Textures”, in 41st Annual IEEE International Carnahan Conference on Security Technology, ISBN: 1-4244-1129-7, pp. 52-58, Ottawa, Canada, 8-11 October 2007.

The signatures are in "jpg" format, 256 gray levels and 120 dpi of resolution. The files of the hands are named xxx\manoxxx_yy.jpg where xxx is the number of the signer and yy its repetition. The palm images are given in the files named xxx\palmaxxx_yy.jpg. The hand contour and landmarks are given in the Matlab2007 file metadata.mat. How to use these files can be seen in the file ReadDatabase.m.

LICENSE AGREEMENT FOR NON-COMMERCIAL RESEARCH USE OF GPDS150hand CORPUS

 

 

 

   

 

 

GPDS hand 3 Band

GPDS100hand3Band database

 

Our hands database consists of 10 times 3 acquisitions (visible, 850nm and 1470nm bands) from 100 people. The 3000 images were taken from the users’ right hand. Most of the users are between 23 to 40 years old. Approximately half of the database volunteers are male. The user in our system can place the hand palm freely over the plate; pegs, templates or any other annoying method for the users to capture their hands are not used. The cameras acquire the hand dorsum image. The image in the 1470nm band is acquired by a XENICS camera XEVA 1.7-320 with an InGaAs sensor, sensitive from 900 to 1700nm, with a band pass filter lens centered on 1470nm and bandwidth of 250nm. The image in the visible band is acquired with a color webcam quickcam E2500, with a resolution of 640x480 pixels. The image in the NIR band is acquired with a color webcam quickcam E2500, with a resolution of 640x480 pixels and a high pass filter lens with cutoff wavelength at 850nm. The procedure is described in:

Miguel A. Ferrer, Aythami Morales, “Hand Shape Biometrics combining the visible and Short Wave In fraReds Bands”, in IEEE Transactions on Information Forensics & Security, Accepted.

The signatures are in "bmp" format as given by the cameras. The files of the hands are named yyy\xxx_yyy_rr.bmp where xxx is the band (xxx= ‘vis’, 850 or 1450), yyy is the user number and rr its repetition.

 

LICENSE AGREEMENT FOR NON-COMMERCIAL RESEARCH USE OF GPDS hand 3 band

 

 

 

   

 

 

GPDS contactless hands

GPDS100Contactlesshands2Band database

 

Our contactless hands database consists of 10 times 2 acquisitions (visible and 850nm) from 100 people. The 2000 images were taken from the users’ right hand. Most of the users are between 23 to 40 years old. Approximately half of the database volunteers are male. The user places his or her hand over the camera and touchless adjusts the position and pose of the hand in order to overlap with the hand mask drawn on the device screen. When the hand and mask overlap more than 70%, the device automatically acquires both the IR and visible image. Detail can be seen in:

Miguel A. Ferrer, Francisco Vargas, Aythami Morales, “BiSpectral Contactless hand based biometric system”, in 2nd National Conference on Telecommunications (CONATEL 2011), Arequipa, Perú, 17-20 May, 2011.

The acquisition device used consists of two inexpensive, standard web cams that obtain images of the hand at the same time. The so called infrared (IR) webcam acquires images in the infrared band (750 to 1000nm) and the so called visible (V) camera acquires images in the visible range (400 to 700nm). The IR webcam was created by simply taking out the webcam lens that eliminates the infrared radiation and adding a filter that eliminates the visible band. We used Kodak filter No 87 FS4-518 and No 87c FS4-519 with no transmittance below 750 nm.

            The infrared illumination is composed of a set of 24 GaAs infrared emitting diode (CQY 99) with a peak wavelength emis-sion of 925 nm and a spectral bandwidth of 40 nm. The diodes were placed in an inverted U shape with the IR and V webcams in the middle. The open part of the U shape will coincide with the wrists of the hand image. The focus of the IR webcam lens is adjusted manually the first time the webcam is used.

 

The signatures are in "bmp" format as given by the webcam. The files of the hands in visible band and infrared band are named xxx\visible_xxx_yy.bmp and xxx\Infraro_xxx_yy.bmp where xxx is the number of the signer and yy its repetition. The segmented palm images are given in the files named xxx\palma_xxx_yy.jpg and the contour of the visible image is given in the files xxx\Icontvisible_xxx_yy. How to use these files can be seen in the file ReadDatabase.m.

 

LICENSE AGREEMENT FOR NON-COMMERCIAL RESEARCH USE OF GPDS Contactless hands

 

 

 

 

   

 

 

GPDS handSWIR hyperspectral database

Sample of GPDShandSWIRhyperspectral database

 

The GPDShandsSWIRhyperspectral database consists of 10 different samples from154 people. Each sample is composed of 350 images acquired by the hyperspectral device, so the total number of images on the database is  images in 256 bands between 900nm and 1600nm. The users were allowed to wear wristwatches or wristlets. The age of the users ranged from 18 to 60 years, 86% of them between 18 and 30. Approximately 70% are students and teachers from our university and the remaining 30% are administration and cleaning staff. In the database, 64% of the users are male. For the first 24 users, we acquired the hyperspectral images from each hand side, palm and dorsum.

 

More information can be found at:

 

Miguel A. Ferrer, Aythami Morales and Alba Díaz, “An approach to SWIR Hyperspectral Hand Biometrics”, accepted at Information Science, doi: 10.1016/j.ins.2013.10.011.

 

The acquisition device used consist on a Xenics® Xeva-1.7-320 camera which is based on an InGaAs detector, sensitive from 900 to 1700nm. The camera provides 256 gray level images with a resolution of 320 by 256 pixels. We used this in conjunction with a SPECIM® Imspector N17E optical spectrograph with numerical aperture f/2.0. This transforms the SWIR camera into a line spectral imaging device, as seen in Figure 2 which shows the reflectance along a longitudinal line (x axis) that crosses four fingers. The spectrographic images consist of 320 pixels in the x axis and 256 bands in the wavelength axis. As the aperture is f/2.0 and the distance from the lens to the plate is 41 cm, the x dimension covered is 21 cm. Therefore the horizontal resolution is approximately 38 dpi.

To add the information for the y axis, a rotating mirror scanner is attached to the objective lens of the spectrograph. As the mirror scanner rotates through 40º, the y spatial dimension covered is 29 cm.

The acquisition procedure is as follows. The user is asked to place the hand dorsum over the plate with the hand palm up; we therefore acquire the hyperspectral image of the hand palm. Once the hand is still, the mirror starts to rotate from minus 20 to plus 20 degrees. The camera acquires around 12 pictures per second. As the procedure takes 30 seconds approximately, we have 350 images in the y axis, with a vertical resolution of 30 dpi approximately. A reconstruction stage makes the hyperspectral cube a coherent object.

The XEVA 1.7-320 had a pixel operability of 99%. It produced 298 erroneous pixels the values of which we interpolated as the mean of the content of the pixels immediately above and below. We did not use the average of all the surrounding pixels because of row contiguous pixel errors.

The illumination system consists of two tungsten filament bulbs with a radiation spectrum from 400 to 1600nm. Most of the bulb energy is centered in the NIR. As can be seen in Figure 2, the bulbs are situated at approximately the same distance from each side of the hand, to avoid shadows or unbalanced lighting.

 

LICENSE AGREEMENT FOR NON-COMMERCIAL RESEARCH USE OF GPDShandSWIRhyperspectral database

 

 

 

 

   

 

 

GPDS Veins

GPDS100VeinsCCDcylindrical database

 

The database consists of 10 different acquisitions of 102 people. The samples were acquired in two separated session one week: five the first time and other five samples the second session. The 1020 images have been taken from the users’ right hand. The system to capture near infrared images of the hand dorsum consists of two arrays of 64 LEDs in the band of 850nm, a CCD gigabit Ethernet PULNIX TM3275 camera with a high pass IR filter with 750nm as cutoff frequency, and a handle with two pegs for positional reference as described in:

 

Miguel A. Ferrer, Aythami Morales, Lourdes Ortega, “Infrarred hand dorsum images for identification”, in Electronic Letters, vol 45, No 6, pp. 306-308, ISSN 0013-5194, 12 March 2009.

 

The signatures are in "bmp" format as given by the camera. The files of the hand veins are named xxx\mano-xxx-yyy.bmp where xxx is the number of the signer and yyy its repetition. A readdatabase.m file is provided.

 

 

GPDS100VeinsCMOScylindrical database

 

The database consists of 10 different acquisitions of 103 people. The samples were acquired in two separated session one week: five the first time and other five samples the second session. The 1030 images have been taken from the users’ right hand. The system to capture near infrared images of the hand dorsum consists of two arrays of 64 LEDs in the band of 850nm and a CMOS webcam with a high pass IR filter with 750nm as cutoff frequency, and a cylindrical handle with two pegs for positional reference.

 

The users of VeinsCMOScylindrical and CMOSergonimic database are the same.

 

The signatures are in "bmp" format as given by the camera. The files of the hand veins are named xxx\venas_xxx_yy.bmp where xxx is the number of the signer and yy its repetition. A readdatabase.m file is provided.

 

 

GPDS100VeinsCMOSergonomic database

 

The database consists of 10 different acquisitions of 103 people. The samples were acquired in two separated session one week: five the first time and other five samples the second session. The 1030 images have been taken from the users’ right hand. The system to capture near infrared images of the hand dorsum consists of two arrays of 64 LEDs in the band of 850nm and a CMOS webcam with a high pass IR filter with 750nm as cutoff frequency, and an ergonomic handle which fix the hand position in a suitable way for the user.

 

The users of VeinsCMOScylindrical and CMOSergonimic database are the same.

 

The signatures are in "bmp" format as given by the camera. The files of the hand veins are named xxx\venas_xxx_yy.bmp where xxx is the number of the signer and yy its repetition. A readdatabase.m file is provided.

 

LICENSE AGREEMENT FOR NON-COMMERCIAL RESEARCH USE OF GPDS Veins

 

 

   

 

 

GPDS Lips

GPDS50Lips database

 

The database consists of 10 different acquisitions of the face of 51 people looking for highlight his/her lips with a CCD camera as described in:

 

Carlos. M. Travieso, J. Zhang, P- Miller, Jesús B. Alonso, Miguel A. Ferrer, “Bimodal biometric verification based on face and lips”, Neurocomputing, ISSN 0925-2312 vol. 74, pp. 2407-2410, 2nd June 2011

 

The pictures are in "jpg" format. The files are named xxx\labio-xx-yy.jpg where xx is the number of the signer and yy its repetition.

 

A file in Matlab to read the database is provided: ReadDatabase.m

 

 

 

   

 

 

gpds SIGNATURE

CORPORA

GPDSsyntheticSignature database

 

Off line signature database. It contains data from 4000 synthetic individuals: 24 genuine signatures for each individual, plus 30 forgeries of his/her signature. All the signatures were generated with different modeled pens. This database replaces previous signatures databases.

 

The synthetic users have been generated following the procedure described at:

 

Miguel A. Ferrer, Moises Diaz-Cabrera and Aythami Morales, “Static Signature Generation: A Neuromotor Inspired Approach”, submitted to IEEE Transaction on Pattern Analysis and Machine Intelligence, October 2013.

Meanwhile, if you are using the database, we ask you reference:

Miguel A. Ferrer, Moisés Díaz, Aythami Morales , “Synthetic Off-line Signature Image Generation”, in Proceedings of 6th IAPR International Conference on Biometrics, Madrid, Spain, 4-6 June 2013.

 

The signatures are in "jpg" format and equivalent resolution of 600 dpi. The files of the genuine signatures are named xxx\c-xxx-yy.png and the files of the forgeries are named xxx\cf-xxx-yy.png where xxx is the number of the signer and yy its repetition.

 

For performance reference, our results with previous public databases using the verifier used at:

 

Miguel A. Ferrer, Francisco Vargas, Aythami Morales, Aaron Ordoñez, "Robustness of Off-line Signature Verification based on Gray Level Features", IEEE Transactions on Information Forensics and Security, vol. 7, no. 3, pp. 966-977, June 2012.

 

Are the next

Training 10 samples

Random Forgeries

Simulated Forgeries

Database

Users

MCYT

75

0.35 %

11.54 %

GPDS960

75

0.37%

13.78%

GPDS960

150

0.44%

15.90%

GPDS960

881

0.88%

23.42%

 

The results with the synthetic database are the next, training with 10 genuine samples

Number of users

Random Forgeries

Simulated Forgeries

75

0.82%

16.01%

150

1.05%

16.44%

881

1.26%

15.19%

1500

1.15%

15.08%

2500

1.10%

15.13%

4000

1.08%

15.09%

 LICENSE AGREEMENT FOR NON-COMMERCIAL RESEARCH USE OF GPDSsynthetic Signature CORPUS

GPDS960signature database

 

 

Off line signature database. It contains data from 960 individuals: 24 genuine signatures for each individual, plus 30 forgeries of his/her signature. The 24 genuine specimens of each signer were collected in a single day writing sessions. The forgeries were produced from the static image of the genuine signature. Each forger was allowed to practice the signature for as long as s/he wishes. Each forger imitated 3 signatures of 5 signers in a single day writing session. The genuine signatures shown to each forger are chosen randomly from the 24 genuine ones. Therefore for each genuine signature there are 30 skilled forgeries made by 10 forgers from 10 different genuine specimens.

 

The signatures are in "bmp" format, in black and white and 300 dpi. The files of the genuine signatures are named xxx\c-xxx-yy.bmp and the files of the forgeries are named xxx\cf-xxx-yy.bmp where xxx is the number of the signer and yy its repetition

 

As the background of the scanned signatures is well contrasted with the darker signature strokes, the signature images where binarized by thresholding with a fixed threshold and  a sort of hair sticking out from signature strokes was eliminated [1]

 

A file in Matlab to read the database is provided: ReadDatabase.m

 

 

[1] M. Blumenstein, Miguel A. Ferrer, J.F. Vargas, “The 4NSigComp2010 off-line signature verification competition: Scenario 2”, in proceedings of 12th International Conference on Frontiers in Handwriting Recognition, ISSBN: 978-0-7695-4221-8, pp. 721-726, Kolkata, India, 16-18 November 2010.

 

 

 

4NSigComp2010 Scenario 2

Off-line signature verification competition database

 

 

The off line signature verification competition held during the 12th International Conference on Frontiers in Handwriting Recognition (ICFHR 2010, Kolkata, India) used as database  a subset of the GPDS960Signature database.

As training subset, the 4NSigCom2010 used 4 genuine signatures of the individuals 301 to 700 of the GPDS960signature corpus. The files of the genuine signatures are named Trainingset\xxx\c-xxx-yy.bmp being xxx the id of the signer which goes from 301 to 700 and yy the repetition from 01 to 04

A Matlab scrip to read and display the train images is provided: ReadTrainingSignatures.m

For testing, 30000 questioned signature images obtained from the GPDS960signature database were used. The test data includes original signatures of GPDS960signature signers 301 to 700, simulated forgeries of each user and random signatures from users 701 to 960. The test files has been named c-xxxxx-yyy.bmp being xxxxx the number of file from 00001 to 30000 and yyy the id of the signer identity claimed from 301 to 700.

A Matlab scrip to read and display the trest images is provided: ReadTestSigantures.m

To evaluate your automatic signature verifier with the 4NSigComp2010 Scenario 2 database, a matlab program called EvaluateASV is provided. As it has been done, this program needs the program asv.m and the file “4nSigCompSignatureIdentification.mat” which contains the matrix called sign.

The program asv.m should be a matlab function defined as:

Function decision=asv(signature,id)

Where signature is the image of the signature to be verified and id is the identity claimed. Decision is supposed 1 if the signature is accepted as genuine and 0 if the signature is considered a forgery.

The mean of the sign matrix values are the next: The signature c-xxxx-yyy.bmp of the test set corresponds to the repetition sign(xxxx,2) of the signer sign(xxxx,1) in the GPDS 960Signature database; The yyy identity claimed by signature c-xxxx-yyy.bmp is equal to sign(xxxx,3). Finally, if sign(xxxx,4) is equal to 0 means that c-xxxx-yy.bmp is a genuine signature; if sign(xxxx,4) is equal to 1 means that c-xxxx-yyy.bmp is a imitation or simulated forgery while if sign(xxxx,4) is equal to 2 means that c-xxxx-yyy.bmp corresponds to random forgery.

For further information, please read [1].

[1] M. Blumenstein, Miguel A. Ferrer, J.F. Vargas, “The 4NSigComp2010 off-line signature verification competition: Scenario 2”, in proceedings of 12th International Conference on Frontiers in Handwriting Recognition, ISSBN: 978-0-7695-4221-8, pp. 721-726, Kolkata, India, 16-18 November 2010.

 

 

 

 

GPDS960GRAYsignature database

This database contains a gray level version of genuine signatures and imitations of the GPDS960Signature database.

 

Due to a move, unfortunately we lost the signatures of 79 users and 143 imitations of the remainder signers. So, the GPDS960GRAYsignature database consists of 881 users, 21144 genuine signatures and 26317 imitations. Total: 47485 signatures.

 

 The lost users and imitations are specified in the ReadDatabase.m file.

 

 In this case, the signatures are in "png" format and have been scanned at 600 dpi. The files of the genuine signatures are named xxx\c-xxx-yy.png and the files of the forgeries are named xxx\cf-xxx-yy.png where xxx is the number of the signer and yy its repetition.

 

 This version of the data base has been obtained scanning the sheets again at 600dpi. So, the segmentation errors are supposed different to those of GPDS960Signature database

 

 This database has been used in:

 

 Miguel A. Ferrer, Francisco Vargas, Aythami Morales, Aaron Ordoñez, "Robustness of Off-line Signature Verification based on Gray Level Features", IEEE Transactions on Information Forensics and Security, vol. 7, no. 3, pp. 966-977, June 2012.

 

 

 

Checks database

 

This database contains 20 images: 12 bank checks and 8 invoices with varying degrees of background complexity.

 

This database has been used in the paper:

 

Miguel A. Ferrer, Francisco Vargas, Aythami Morales, Aaron Ordoñez, "Robustness of Off-line Signature Verification based on Gray Level Features", IEEE Transactions on Information Forensics and Security, vol. 7, no. 3, pp. 966-977, June 2012.

 

Please, refer to it if you use this database.

 

The submitted paper blended the MCYT (http://atvs.ii.uam.es/databases.jsp ) and GPDS960GRAYSignatures database with the check database to obtain a synthetic signature database with distorted gray levels. We used the multiply blend mode which multiplies the check image by the signature one. As we overlay gray level strokes, each stroke results in a new darker gray level.

 

The next Matlab files are provided:

 

1.      Read the check database divided in three gray level distortions as in the submitted paper: ReadCheckDatabase.m.

2.     Blend a signature with a given signature: BlendSignaturewithCheck.m

3.     As we have used texture parameters based on Local Binary Patterns (LB), Local Directional Pattern (LDP) and Local Derivative Pattern (LDerivP), the programs to work them out are also provided in the next Matlab files: LBP.m, LDP.m and LDeriv.m

 

CHECKS DATABASE

 

 

 

 

grupo de procesado digital de señales
departamento de señales y comunicaciones
Campus de Tafira, 35017 - Las Palmas

Phone: +34 928 451 269   fax: +34 928 451 243
email: gpds@gi.ulpgc.es