Abstract

BACKGROUND
This description relates to rescoring text hypotheses in text recognition based on segmental features. Offline printed text and handwriting recognition (OHR) can be a challenging research problem for many reasons. In many recognition approaches, segmentation of handwritten text is inaccurate because of stylistic variations in connected Scripts. Also, images suffer degradations that result in breaks and merges in glyphs, which creates new connected compo nents that are not accurately recognized by classifiers. Statis tical approaches have been developed that do not rely on segmentation, but such systems lack the use of segmental features.

Date

2014

Authors

P Natarajan, R Prasad, R Schwartz, K Subramanian

Inventors

Premkumar Natarajan, Rohit Prasad, Richard Schwartz, Krishnakumar Subramanian

Patent_office

US

Patent_number

8644611

Application_number

12477582