|
|
Evaluation:
The
ICDAR 2013 Handwriting Segmentation Contest will be based on the
evaluation strategies of previous document segmentation contests (ICDAR
2007 Handwriting Segmentation Contest, ICDAR
2009 Handwriting Segmentation Contest and ICFHR
2010 Handwriting Segmentation Contest).
- The
performance evaluation method will be based on counting the number of
one-to-one matches between the areas detected by the algorithm and the areas in
the ground truth. We will use a MatchScore table whose values will be
calculated according to the intersection of the ON pixel sets of the
result and the ground truth.
- Let
Gi the
set of all points of the
i ground truth region, Rj the set of all points of the j result region,
Τ(s)
a function that counts the elements of set s. Table
MatchScore(i,j)
represents the matching results of i ground truth
region and the j result region as follows: MatchScore(i,j)=T(Gi
 Rj)/T(Gi  Rj).
For example:
For a given ground truth:
and a segmentation result:
MatchScore
is calculated as follows:
e.g. MatchScore (1,5) = T(G1 R5)/T(G1 R5) = 11/14 = 0.79
- Detection
Rate and Recognition
Accuracy are calculated from the MatchScore table1.
1 I.
Phillips and A. Chhabra, "Empirical Performance Evaluation of Graphics
Recognition Systems," IEEE Transaction of Pattern Analysis and Machine
Intelligence, Vol. 21, No. 9, pp. 849-870, September 1999.
|
|
|
|
