Difference between revisions of "Character discovery in the sub-word shapes"
(→Description) |
|||
| Line 1: | Line 1: | ||
| − | [[Datasets]] -> Current Page | + | [[Datasets]] -> [[Datasets List]] -> Current Page |
{| style="width: 100%" | {| style="width: 100%" | ||
Revision as of 18:06, 27 January 2011
Datasets -> Datasets List -> Current Page
|
Description
Labels for 15 characters are provided in the ground truth. For each character, a classifier is required to predict the presence of that character in each shape. Output of the classifier is binary. The evaluation for each character is carried out separately. The Balanced Error Rate (BER) is used as the performance measure (please see below for the details).
As a reference, the results can be compared to the published results available in Table 2 in [1]. Evaluation Protocol A cross-validation technique is proposed for the evaluation of this task. The average BER for each character is computed by repeating the training process for 10 times. In each run, the database is split into a training set and test set randomly. The training set consists of 80 percent of the database. The proposed method is trained using the training data, and its performance is computed over the test data in terms of BER.
The BER is defined as:
BER = 0.5*(FP/(TN+FP) + FN/(FN+TP))
Where,
FP = False Positive TP = True Positive FN = False Negative TN = True Negative
The average BER is calculated over the 10 runs.
Related Dataset
Related Ground Truth Data
References
- Reza Farrahi Moghaddam, Mohamed Cheriet, Mathias M. Adankon, Kostyantyn Filonenko, and Robert Wisnovsky, “IBN SINA: A database for research on processing and understanding of Arabic manuscripts images”, Proceedings of DAS’10, June 9-11, 2010, Boston, MA, USA
This page is editable only by TC11 Officers .