© Springer-Verlag London 2014. All rights are reserved. The definition of standard frameworks for performance evaluation is a key issue in order to advance the state-of-the-art in any field of document analysis since it permits a fair and objective comparison of different proposed methods under a common scenario. For that reason, a large number of public datasets have emerged in the last years. However, several challenges must be considered when creating such datasets in order to get a sufficiently large collection of representative data that can be easily exploited by the researchers. In this chapter we review different approaches followed by the document analysis community to address some of these challenges, such as the collection of representative data, its annotation with ground-truth information, or the representation using accepted and common formats. We also provide a comprehensive list of existing public datasets for each of the different areas of document analysis.
|Title of host publication||Handbook of Document Image Processing and Recognition|
|Number of pages||26|
|Publication status||Published - 1 Jan 2014|
- Generation of synthetic data
- Performance evaluation
- Public datasets