Multilingual Multimodal Biometric Database (MPEG2 compressed version) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Description ~~~~~~~~~~~ The database is composed of: * signatures * video: + audio + still pictures Audio and still pictures are generated from the video files with perl scripts and mplayer (http://www.mplayerhq.hu/) software. The size of the database without extraction of audio and still pictures is 1801 MB and 8078 MB after the extraction. If you don't need audio or video, you can comment the execution of the generating screept by editing the Makefile. For the impatient: ~~~~~~~~~~~~~~~~~~ Prerequisites: make, perl, mplayer, sox Run: $ make to generate the pictures and audio files Signatures: ~~~~~~~~~~~ Fake or modified signatures were collected and saved in files with .dat extension. Singature files are composed of comma separated integer values, with: X,Y,pressure,time denoted X,Y,P,T from now on. X can take values in the interval of 0 to 10207. Y can take values in the interval of 0 to 7423. P can take values in the interval of 0 to 511. T is measured in ms and always starts on 0. Pen ups are denoted with X=-1,Y=-1,P=-1, and arbitrary T value, take this into account when building statistical models, because X=-1,Y=-1 is not a "real" spatial coordinate of the signature. * Capturing Device: Wacom "Graphire3 4x5" digitizer tablet. Sensing surface 127.6mm x 92.8mm. Spatial reolution is 2032 lpi (lines per inch). 512 different pressure levels. Sampling rate is about 100Hz (varying). http://www.wacom.com/graphire/4x5.cfm Audio and Video ~~~~~~~~~~~~~~~ The videos provide audio and still pictures. Video is encoded raw UYVY. AVI 640 x 480, 15.00 fps with uncompressed 16bit PCM audio; mono, 32000 Hz. A few videos are with uncompressed PCM audio; stereo, 44100 Hz little endian. Audio: ~~~~~~ Audio is extracted as 16 bit PCM WAV file (with wav header), sampled at 16000 Hz, mono little endian. BUGS: there was a problem with mplayer not writting the byte_alignment of the wav header correctly, which caused files not being read correctly on Matlab. A patch fixing the bug was sent and merged in the mplayer CVS. We include the patch with the database. Against CVS revision: 1.29 of /cvsroot/mplayer/main/libao2/ao_pcm.c The patch should also work against MPlayer-1.0pre7, which was the latest official release of mplayer which was unpatched. So using the CVS version is recommended, until a patched official release is made. If not, --fixheader option can be used for separate audio from video stream* scripts. * Capturing Device: Shure SM58 microphone. Frequency response 50 Hz to 15000 Hz. Unidirectional (Cardiod) dynamic vocal microphone. Still Pictures ~~~~~~~~~~~~~~ There are perl scripts for extraction of audio and still pictures from the videos, extraction of audio takes significantly less time than picture extraction. Uncompressed PNG (portable network graphics) files are extracted from the video files. * Capturing Device: Allied Vision Technologies AVT marlin MF-046C 10 bit ADC, 1/2" (8mm) Progressive scan SONY IT CCD http://www.alliedvisiontec.com/indexc.php?p=91&t=produktinfos&o=3&a=sp