Introduction

There are 19 tools for measuring the performance and quality of OCR output. See the user guide (user-guide.pdf) for more information.

Building

To build the tools, run:

make

To install them, run:

make install

To install them to a different directory, run:

make PREFIX=/my/install/directory install

Running tests with UTF-8

Run the tool inside the ocrevalutf8 wrapper script, e.g.

ocrevalutf8 accuracy correct.txt ocr.txt or ocrevalutf8 wordacc correct.txt ocr.txt

This is needed as the tools don't (yet) natively handle UTF-8, but they can handle strangely encoded UTF-16, which this wrapper script delivers.

Testing Tesseract with a .traineddata file

'tessaccsummary' is an included script to make testing the accuracy of a Tesseract training file simple and easy.

The script expects a directory laid out as follows:

images/ images/firstimg.png images/firstimg.txt images/secondimg.png images/secondimg.txt

The .txt file should be a UTF-8 file of the correct text ("ground truth") for the .png file of the same name. The image and text files may be named anything.

You can then run the script like so:

tessaccsummary images/ /path/to/lang.traineddata

This runs Tesseract using the provided .traineddata file on each .png image in the image directory, and print the character accuracy for each image.

You can instead get the word accuracy for each image like this:

tessaccsummary -w images/ /path/to/lang.traineddata