Abstract:
Text detection in natural images is an important prerequisite for many content-based
image analysis tasks. Although it is widely studied in recent years, due to unpredictable
scene environment, reading texts is still quite challenging and continues to be an open
research problem.
It is recently being shown how the state-of-the-art object detection methods can be
modi ed and then applied successfully for the purpose of scene text detection. A method,
based upon YOLOv2 and RPN, intended to do end-to-end text recognition, achieves
state of the art accuracy in the complete scene text recognition on two standard datasets
ICDAR-2013 and ICDAR-2015, even while working at the real time and being faster than
the other competing methods.
This method is improved upon to give better text detection and localization results,
in real time speeds.
