Text retrieval from early printed books

Reviewer: Sithu D. Sudarsan

Historians and archaeologists address the challenge of reading ancient scriptures. Now, computer scientists are extending recent advances in text retrieval techniques to ancient scriptures. Using the Latin Gutenberg Bible, Marinai's work is on the identification of character objects. Character objects are connected components or portions thereof that correspond to single characters. The retrieval technique has a preprocessing phase in which columns, text lines, and character objects are extracted. Subsequently, character clustering is performed by creating initial vectors and using them to train self-organizing maps; this step is very crucial for the success of the work. Finally, the matching algorithm uses the query-by-example approach. The various steps of the algorithm are nicely described in the paper. Marinai explains the experiments she carried out and provides an outline of the issues she faced. To increase the confidence level in the algorithm, she proposes to continue the experiments. In spite of its interesting aspects, the approach does have its limitations. If various formats are used, the proposed preprocessing may not be feasible for documents. This makes the technique useful for one class of documents, rather than generic. Also, the selection of the initial vector set needs to cover sufficient details to capture as many character objects as possible. These limitations are dependent on subject matter experts, and the field has a long way to go. Online Computing Reviews Service