TriGraphSlant - benchmark set for writer identification - writers were asked to write in unnatural slant

 
Disclaimer and terms of use:
============================
 
/*****************************************************************************\
*                                                                             *
*                                                                             *
*   This is the TrigraphSlant (Img version) Distribution, release 18/3/2011   *                                 *
*                                                                             *
*   This distribution contains 188 images of scanned handwritten text,        *
*   scanned at resolution 300dpi Canon LiDE 25, grey scale,                   *
*   by 47 Dutch writers, four pages per writer, from four                     *
*   writing conditions, one condition per page. The conditions are:           *
*   1. [AN] Copy text A in your natural handwriting.                          *
*   2. [BN] Copy text B in your natural handwriting.                          *
*   3. [BL] Copy text B and slant your handwriting to the                     *
*           left as much as possible.                                         *
*   4. [BR] Copy text B and slant your handwriting to the                     *
*           right as much as possible.                                        *
*   The codes AN, BN, BL and BR refer to subsets into which the collected     *
*   pages of the writers were subdivided. AN represents a collection of       *
*   authentic documents; BN, BL and BR can be seen as collections of          *
*   questioned documents. To avoid structural effects of fatigue, the order   *
*   of item 3 and 4 was randomized at each collection: half of the subjects   *
*   wrote the BR page before the BL page. The data were collected at three    *
*   sites, in three cities: The Hague: NFI (N...), Donders Institute for      *
*   Brain, Cognition and Behaviour, Radboud University Nijmegen (D...)        *
*   and the Artificial Intelligence Dept. of University of Groningen (R...)   *
*                                                                             *
*   Copyright The International Unipen Foundation, 2010, All rights reserved  *
*******************************************************************************
*                                                                             *
*                                                                             *
*  DISCLAIMER AND COPYRIGHT NOTICE FOR ALL DATA CONTAINED ON THIS CARRIER:    *
*                                                                             *
*                                                                             *
*  1) PERMISSION IS HEREBY GRANTED TO USE THE DATA FOR RESEARCH               *
*     PURPOSES. IT IS NOT ALLOWED TO DISTRIBUTE THIS DATA FOR COMMERCIAL      *
*     PURPOSES.                                                               *
*                                                                             *
*                                                                             *
*  2) PROVIDER GIVES NO EXPRESS OR IMPLIED WARRANTY OF ANY KIND AND ANY       *
*     IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR PURPOSE ARE       *
*     DISCLAIMED.                                                             *
*                                                                             *
*  3) PROVIDER SHALL NOT BE LIABLE FOR ANY DIRECT, INDIRECT, SPECIAL,         *
*     INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF ANY USE OF THIS      *
*     DATA.                                                                   *
*                                                                             *
*  4) THE USER SHOULD REFER TO THE FOLLOWING ARTICLE ON THIS DATA SET:        *
*                                                                             *
* A.A. Brink, R.M.J. Niels, R.A. van Batenburg, C.E. van den Heuvel,          *
* L.R.B. Schomaker, Towards robust writer verification by correcting          *
* unnatural slant, Pattern Recognition Letters, Volume 32, Issue 3,           *
* 1 February 2011, Pages 449-457, ISSN 0167-8655,                             *
* DOI: 10.1016/j.patrec.2010.10.010.                                          *
*                                                                             *
*  5) THE RECIPIENT SHOULD REFRAIN FROM PROLIFERATING THE DATA SET TO THIRD   *
*  PARTIES EXTERNAL TO HIS/HER LOCAL RESEARCH GROUP. PLEASE REFER INTERESTED  *
*  RESEARCHERS TO HTTP://UNIPEN.ORG FOR OBTAINING THEIR OWN COPY.             *
\*****************************************************************************/
 
Abstract
 
Towards robust writer verification by correcting unnatural slant
 
A.A. Brink, , R.M.J. Niels, R.A. van Batenburg, C.E. van den Heuvel,   
and L.R.B. Schomaker,  
 
a Institute of Artificial Intelligence and Cognitive Engineering (ALICE),  
  University of Groningen, P.O. Box 407, 9700 AK Groningen, The Netherlands
   
b Donders Institute for Brain, Cognition and Behaviour, Radboud University Nijmegen,  
  P.O. Box 9104, 6500 HE Nijmegen, The Netherlands
   
c Netherlands Forensic Institute, P.O. Box 24044, 2490 AA Den Haag, The Netherlands
 
Received 11 September 2009.  Available online 30 October 2010.
 
Slant is a salient feature of Western handwriting and it is considered to be an
important writer-specific feature. In disguised handwriting however, slant is
often modified. It was tested whether slant is indeed an important factor and it
was tested whether the distorting effect of deliberate slant change can be
countered by a simple shear transform. This was done in two off-line writer
verification experiments in image processing conditions of slant elimination and
slant correction. The experiments were performed using three features based on
statistical pattern recognition, including the state-of-the-art features
Fraglets and Hinge. A new public dataset was created and used, containing
natural and slanted handwriting by 47 writers. A striking result is that the
average natural slant value is much less important for biometric systems than is
usually assumed: eliminating slant yields just a 1-5% performance loss. A
second result is that the effects of deliberate slant change cannot be fully
countered by a simple shear transform: it raises performance on the distorted
handwriting from 53-68% to 64-90%, but this is still lower than normal
operation on natural handwriting: 97-100%.
 
Research highlights
- The value of slant as a writer identification feature has been overrated.  
- Deliberate slant change can be partly countered by the shear transform.  
- Deliberate slant change introduces non-affine distortions to the handwriting.  
- A new dataset of deliberately slanted handwriting was introduced.
 
Keywords: Handwriting biometrics; Writer verification; Slant; Disguise; Statistical
pattern recognition