Refine
Year of publication
- 2009 (13) (remove)
Document Type
- Conference Proceeding (6)
- Report (4)
- Working Paper (3)
Has Fulltext
- yes (13)
Is part of the Bibliography
- no (13)
Keywords
- Herbarium (13) (remove)
Die vorliegende Arbeit untersucht den möglichen Einsatz kommerzieller Software im Projekt Herbar Digital. Dabei werden zwei Kategorien unterschieden: OCR-Software und Barcodesoftware. Von der ersten Kategorie gibt es eine Vielzahl käuflicher Programme auf dem Markt sowie auch einige kostenlose Freewareprogramme. Die Qualität ist jedoch sehr unterschiedlich, insbesondere fallen die Freewareprogramme stark ab. Es kristallisieren sich vier hochqualitative Programme heraus, die genau untersucht werden. Von diesen eignen sich zwei für das Projekt, wobei Omnipage 16 der Vorzug gegeben wird. In der Kategorie der Barcodesoftware fiel die Wahl auf QS-Barcode 4.0, da sich OCR-Programme für das Lesen von Barcodes als ungeeignet erwiesen. Die anfängliche Erkennungsrate von 90% konnte durch eigene Verfahren zur Bildvorverarbeitung auf 100% gesteigert werden.
The methods developed in the research project "Herbar Digital" are to help plant taxonomists to master the great amount of material of about 3.5 million dried plants on paper sheets belonging to the Botanic Museum Berlin in Germany. Frequently the collector of the plant is unknown. So a procedure had to be developed in order to determine the writer of the handwriting on the sheet. In the present work the static character is transformed into a dynamic form. This is done with the model of an inert ball which is rolled through the written character. During this off-line writer recognition, different mathematical procedures are used such as the reproduction of the write line of individual characters by Legendre polynomials. When only one character is used, a recognition rate of about 40% is obtained. By combining multiple characters, the recognition rate rises considerably and reaches 98.7% with 13 characters and 93 writers (chosen randomly from the international IAM-database [3]). Another approach tries to identify the writer by handwritten words. The word is cut out and transformed into a 6-dimensional time series and compared e.g. by means of DTW-methods. A global statistical approach using the whole handwritten sentences results in a similar recognition rate of more than 98%. By combining the methods, a recognition rate of 99.5% is achieved.
The research project "Herbar Digital" was started in 2007 with the aim to digitize 3.5 million dried plants on paper sheets belonging to the Botanic Museum Berlin in Germany. Frequently the collector of the plant is unknown, so a procedure had to be developed in order to determine the writer of the handwriting on the sheet. In the present work the static character was transformed into a dynamic form. This was done with the model of an inert ball which was rolled along the written character. During this off-line writer recognition, different mathematical procedures were used such as the reproduction of the write line of individual characters by Legendre polynomials. When only one character was used, a recognition rate of about 40% was obtained. By combining multiple characters, the recognition rate rose considerably and reached 98.7% with 13 characters and 93 writers (chosen randomly from the international IAM-database [3]). A global statistical approach using the whole handwritten text resulted in a similar recognition rate. By combining local and global methods, a recognition rate of 99.5% was achieved.