ContactPerson: slavik@cse.buffalo.edu
Remote host: avior.cedar.buffalo.edu
Remote ident: slavik
### Begin Citation ### Do not delete this line ###
%R 2000-04
%U /u0/csestaff/stock/slicedist.ps
%A Slavik, Petr
%T Slice Distance
%D April 08, 2000
%I Department of Computer Science and Engineering, SUNY Buffalo
%K handwriting recognition, minimum edit distance, slice distance, dynamic programming, word recognition
%Y I.5.4
%X We introduce a novel way of computing an image-independent 
recognizer-dependent distance between ASCII words. This distance 
is a natural generalization of the minimum edit distance in the  
context of handwriting recognition.  It is based on confusion  
matrices for only parts of characters called character ``slices''. 
This ``slice distance'' naturally explains the confusions and 
misrecognitions encountered in recognition of handwritten words 
and phrases by a particular recognizer and could be used to  
exploit both the weak and the strong points of the recognizer.   
Even though we describe the techniques for computing the slice  
distance using an example of a particular word recognizer, our 
methods can be easily generalized to almost any segmentation-based  
handwritten-word recognition system.