Universität zu Köln

Diem, Markus, Sablatnig, Robert, Gau, Melanie and Miklas, Heinz (2011). Recognizing Degraded Handwritten Characters. In: Kodikologie und Paläographie im digitalen Zeitalter 2 - Codicology and Palaeography in the Digital Age 2, pp. 295-306. Norderstedt: Books on Demand (BoD). ISBN 978-3-8423-5032-8

Preview

PDF 17_gau.pdf Download (3MB)

Abstract

In this paper, Slavonic manuscripts from the 11th century written in Glagolitic script are investigated. State-of-the-art optical character recognition methods produce poor results for degraded handwritten document images. This is largely due to a lack of suitable results from basic pre-processing steps such as binarization and image segmentation. Therefore, a new, binarization-free approach will be presented that is independent of pre-processing deficiencies. It additionally incorporates local information in order to recognize also fragmented or faded characters. The proposed algorithm consists of two steps: character classification and character localization. Firstly scale invariant feature transform features are extracted and classified using support vector machines. On this basis interest points are clustered according to their spatial information. Then, characters are localized and eventually recognized by a weighted voting scheme of pre-classified local descriptors. Preliminary results show that the proposed system can handle highly degraded manuscript images with background noise, e.g. stains, tears, and faded characters.