API example for user patterns
User patterns can be useful when recognizing ID type of fields which have non-dictionary words but follow specific patterns of alphabets and digits e.g. \A\A\d\d\d\d\A
or \A\A\d\d\d\A
This documentation provides a simple example on how to use the tesseract-ocr API (4.x) in C++ for applying user patterns for improving recognition. It is expected that tesseract-ocr is correctly installed including all dependencies. It is expected the user is familiar with C++, compiling and linking program on their platform.
This is based on an example provided in tesseract-ocr forum and updated for the recent implementation of the feature for tesseract 4.x.
Please note that while this example gets 100% accuracy after user_patterns are applied, that may not always be the case. User patterns (like user dictionaries) are merely applied as a hint while decoding, but not exclusively. Pre-processing the image usually improves the quality of recognition and is recommended.
Requirements
In order to apply user patterns for improving recognition, the following are required.
user patterns file
The user patterns file should contain one pattern per line in UTF-8 format. In choosing which patterns to include please be aware of the fact that providing very generic patterns will make tesseract run slower. Best results may be obtained by having a single pattern in the file.
Details of type of patterns that can be used are given in trie.h.
Example of a user patterns file
Make a text file, and write each pattern on a separate line, with UNIX line endings (line-feed character) and a blank line at the end, e.g.
\A\A\d\d\d\d\A
In the following, let’s assume you named that pattern file path/to/my.patterns
.
config file
For the API, the information about the user patterns file needs to be specified in a config file.
(For the CLI, this works as well, but there is also a direct option for the user patterns file alone.)
Example of a config file
Make a text file, and write user_patterns_file
into it verbatim, followed by the path name in one line, with UNIX line endings (line-feed character) and a blank line at the end, e.g.
user_patterns_file path/to/my.patterns
In the following, let’s assume you named that config file path/to/my.patterns.config
.
CLI Example
From the command line, user patterns can be invoked as follows:
tesseract input.tif output --user-patterns path/to/my.patterns
API Example
Take the following image file (Arial.png
) as input:
The following code uses the above user patterns file and config file on that image file:
#include <tesseract/baseapi.h>
#include <leptonica/allheaders.h>
int main()
{
Pix *image;
char *outText;
char *configs[]={"path/to/my.patterns.config"};
int configs_size = 1;
tesseract::TessBaseAPI *api = new tesseract::TessBaseAPI();
if (api->Init(NULL, "eng", tesseract::OEM_LSTM_ONLY, configs, configs_size, NULL, NULL, false)) {
fprintf(stderr, "Could not initialize tesseract.\n");
exit(1);
}
image = pixRead("Arial.png");
api->SetImage(image);
outText = api->GetUTF8Text();
printf(outText);
api->End();
delete api;
delete [] outText;
pixDestroy(&image);
return 0;
}
Build and run script
#!/bin/bash
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/local/include
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib
g++ -std=c++17 -o my.patterns.api my.patterns.api.cpp -llept -ltesseract
export TESSDATA_PREFIX=~/tessdata_best
./my.patterns.api > Arial-patterns.txt
diff -u Arial-patterns.txt Arial-gt.txt
With the user patterns file for this image, the recognition is 100% correct. Without it, there are a number of errors:
--- Arial-patterns-no.txt 2019-07-05 04:21:04.367188492 +0000
+++ Arial-gt.txt 2019-07-05 04:05:11.000000000 +0000
@@ -1,20 +1,20 @@
DQ2679M
LO6217I
QK2101G
-JBO363H
+JB0363H
KN2873M
-Z2B0929J
+ZB0929J
JF3829W
-YNO0584J
-SVv8400Q
+YN0584J
+SV8400Q
FY4523X
KS0016J
OB3016R
VA4335P
-QHO205V
-UH20932
+QH0205V
+UH2093Z
GW3760Y
-S02306T
+SO2306T
XT8204F
-MR6804|
-0OX5866M
+MR6804I
+OX5866M