basic_usage

tesseract 4.0.0 alpha

OCR Engine modes:

• Original Tesseract only.
• Neural nets LSTM only.
• Tesseract + LSTM.
• Default, based on what is available.

Install at windows

• Downoload zip from 4.0.0-alpha for Windows
• unzip to dir tesseract-4.0.0-alpha
• Download language data from here. (chi_sim.traineddata for Simplified Chinese , eng.traineddata for English).
• osd.traineddata is necessity
• mkdir tesseract-4.0.0-alpha/tessdata
• set TESSDATA_PREFIX environment variable to the parent directory of your "tessdata" directory
• Copy *.traineddata to tesseract-4.0.0-alpha/tessdata

Run tesseract

Simple OCR

• English text recognition from test_image2.png

\tesseract-4.0.0-alpha>tesseract.exe test_image2.png 1
Tesseract Open Source OCR Engine v4.00.00alpha with Leptonica

Then we get the recoginized english test at 1.txt

Simple OCR for Chinese

• Chinese text recognition from test_image3.png

tesseract-4.0.0-alpha>tesseract.exe test_image3.png out -l chi_sim
Tesseract Open Source OCR Engine v4.00.00alpha with Leptonica

Then we get the recoginized Chinese test at 1.txt

Get character location

• copy configs/hocr to tesseract-4.0.0-alpha/tessdata/configs/hocr

• run cmd below to get the hocr file

\tesseract-4.0.0-alpha>tesseract.exe test_image2.png out hocr
Tesseract Open Source OCR Engine v4.00.00alpha with Leptonica
OSD: Weak margin (1.71) for 78 blob text block, but using orientation anyway: 0

• then we get a file out.hocr (each word's location is in this file)
• renam out.hocr to out.xml (hocr is different from page xml format, this convertiong will loss some information, but we can still get text region visualization directly )
• download PAGEViewer from http://www.prima.cse.salford.ac.uk/tools/PAGEViewer
• open out.xml and test_image2.png by PAGEViewer, so we get below pic