US 12,380,720 B2
Method, apparatus, and system for character recognition using context
Ting Xu, Campbell, CA (US)
Assigned to KONICA MINOLTA BUSINESS SOLUTIONS U.S.A., INC., Foster City, CA (US)
Filed by KONICA MINOLTA BUSINESS SOLUTIONS U.S.A., INC., Foster City, CA (US)
Filed on Dec. 30, 2022, as Appl. No. 18/092,005.
Prior Publication US 2024/0221408 A1, Jul. 4, 2024
Int. Cl. G06V 30/00 (2022.01); G06V 10/82 (2022.01); G06V 30/19 (2022.01)
CPC G06V 30/19173 (2022.01) [G06V 10/82 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
responsive to an input scanned image, transcribing text from the image into a plurality of aggregated output classes;
performing neural network based fuzzy optical character recognition (OCR), providing fuzzy OCR output comprising a further plurality of aggregated output classes of visually similar characters, and identifying a most probable aggregated output class from the plurality of aggregated output classes;
using a neural network-based language model and the identified most probable aggregated output class, determining an amount of context in which to examine the identified most probable aggregated output class;
responsive to the determined amount of context, examining the context to determine which of the visually similar characters in the identified most probable aggregated output class should be output; and
responsive to the examining, outputting a selected one of the visually similar characters as recognized text.