检测
Text instance level:
Anchor-based methods
EAST
Region proposal methods
R2CNN
Componets-level:
SegLink , Corner localization, CPTN
Pixel-level:
PixelLink
Multi-Oriented Text:
Instance Transformation Network (ITN)
Text of Irregular Shapes:
TextSnake
识别
CTC-based Methods
(a) CNN + softmax. (b) RNN + CTC. (c) RNN +Attention. (d) CNN + CTC
Attention-based methods
FAN
(perspectively distorted or curved)#STN+attention-based Sequence Recognition Network#:The STN predict a Thin-Plate-Spline transformations which rectify the input irregular text image into a more canonical form.
(perspectively distorted or curved)#four feature sequences of four directions#:horizontal, reversed horizontal, vertical and reversed vertical. And a weighting mechanism is designed to combine the four feature sequences.
(perspectively distorted or curved)#alignment loss to regularize the estimated attention at each time-step. Further, they use a coordinate map as a second input to enforce spatial-awareness.
HAM can handle different types of distortion
None End-to-End System
SSD+CRNN
End-to-End
Faster-RCNN + an encoder-decoder based text recognition model
EAST/YOLO2 branch -> text proposals -> map to CTC-based methods
