-
Notifications
You must be signed in to change notification settings - Fork 591
Text Recognition RNN
Philipp Zumstein edited this page Apr 25, 2017
·
2 revisions
Work in Progress Help us by extending these notes...
The command ocropus-rpred
will recognize the text from images of single lines with a RNN.
ocropus-rpred [add here possible options] <input files>
-m MODEL, --model MODEL
line recognition model
-p PAD, --pad PAD extra blank padding to the left and right of text line
-N, --nonormalize don't normalize the textual output from the recognizer
--llocs output LSTM locations for characters
--alocs output aligned LSTM locations for characters
--probabilities output probabilities for each letter
By default a line normalizer is used to dewarp the single lines before the actual recognition:
The dewarping of the text line (first image) tries to find the center (blue curve) and then cut out slices with some fixed radius around the center (cf. this illustration). This will then assure that the center line is horizontally and thereby the text line dewarped.
The default behavior can be influence by these two parameters:
-e, --nolineest deactivate the text normalizer and therefore the dewarping
-l HEIGHT, --height HEIGHT
overrides the target line height, which is 48
These parameters make only sense if the model is also trained under the same conditions and usually there is no need to use them at all.
-r, --estrate estimate error rate only
-c ESTCONF, --estconf ESTCONF
estimate confusion matrix
-C COMPARE, --compare COMPARE
string comparison used for error rate estimate
--context CONTEXT context for error reporting
-s SHOW, --show SHOW if >0, shows recognition output in a window and waits
this many seconds
-S SAVE, --save SAVE save debugging output image as PNG (for bug reporting)
-h, --help show this help message and exit
-n, --nocheck disable error checking on inputs
-q, --quiet turn off most output
-Q PARALLEL, --parallel PARALLEL
number of parallel processes to use, default: 1