作者: Anguelos Nicolaou , Andrew D. Bagdanov , Marcus Liwicki , Dimosthenis Karatzas
DOI: 10.1109/ICDAR.2015.7333855
关键词:
摘要: Sampling Local Binary Patterns, a variant of Patterns (LBP) for text-as-texture classification. By adapting and extending the standard LBP operator to particularities text we get generic classification scheme apply it writer identification. In experiments on CVL ICDAR 2013 datasets, proposed feature-set simple end-to-end pipeline demonstrate State-Of-the-Art (SOA) performance. Among SOA, method is only one that based dense extraction single local feature descriptor. This makes fast applicable at earliest stages in DIA without need segmentation, binarization, or multiple features.