UNIPEN Database Conditions of Use

Two major types of use can be identified:

• I. Non-commercial use

• II. Commercial use

I. Non-commercial use Non-commercial use refers to university and institutional research which aims at public dissemination of research results. This type of usage of UNIPEN data is highly advocated by the International Unipen Foundation (iUF). However, there is a Publication Policy which must be taken into account (See below).

II. Commercial use II.a Commercial use of UNIPEN data proper - the textual content and the point coordinates - is prohibited. An example would be the extraction of handwriting coordinates to sell 'script fonts'. II.b The usage of UNIPEN data for the training of commercial handwriting recognition systems is allowed. II.c The UNIPEN logo will be presented by the user in the final documentation of the resulting software product. II.d Reference to individual writer identities or the identity of individual data donator companies from within the UNIPEN data distribution should be avoided at all times.

Note: Also in the case of commercial development, the user is kindly asked to present the results of the underlying research and development via an acknowledged science & technology forum (journal or conference).

Ad I. UNIPEN Publication Policy

    Reference example: 

        "As a training set, we used UNIPEN [xx] Train-R01/V07, 
         benchmark ..., subsets ..... 
         As a test set, we used UNIPEN DevTest-R01/V02, 
         benchmark ..., subsets .... 
         To the raw UNIPEN data, the following pre-processing 
         was applied: ...."
             .
             .
             .

        [xx] Guyon, I., Schomaker, L., Plamondon, R., 
             Liberman, M. & Janet, S. (1994). 
             UNIPEN project of on-line data exchange and recognizer 
             benchmarks, Proceedings of the 12th International
             Conference on Pattern Recognition, ICPR'94, 
             pp. 29-33, Jerusalem, Israel, October 1994. IAPR-IEEE.

In this example we assume the release of the set DevTest-R01/V02, which will actually take place in the future.

In case your training set and test set are derived from within a single distribution such as Train-R01/V07, please explain in detail how your random selection of samples from within this distribution was produced. Was the process actually random? Was manual pruning involved? Improvements to the labels (truth values) can be submitted by the users in the form of .SEGMENT... entries via email to the iUF.

I.2 - Which data?

Note that there is a problem in the use of test sets. Iterated use of a particular training / test set pair in a development process can be considered as indirect training! Even if a development set as such is not formally used for training, it is a well-known fact that all parameter adjustments, code improvements, etc., are a form of training, regardless of the type of pattern recognition algorithm which is used. Therefore, it is good practice to explain the effort spent in iterated testing in the publications. The tendency to iterate a single training/test set pair within a complete PhD project has led to inflated reported recognition rates in the past. It is good practice to generate a random selection of multiple sets at the start of such projects.

I.3 - Benchmark (eq. database subset) overview