0
$\begingroup$

I want to OCR photos of pages of printed text. Nothing too complicated but the images were taken with smartphone cameras, the lighting is suboptimal and sometimes the images were photocopied before.

So far I successfully used the Tesseract OCR engine with its built-in image binarization. I learned that it uses the Otsu method for binarizing the image. This method has its problems with shaded lighting so I now want to serve pre-binarized images to Tesseract.

While researching which methods could be used for binarizing images I found this paper which compares and ranks different approaches:

Robust Combined Binarization Method of Non-Uniformly Illuminated Document Images for Alphanumerical Character Recognition

by Hubert Michalak and Krzysztof Okarma

(Link)

The top-performing approach incorporates the Structural Symmetry of Strokes (SSP) binarization method which in turn is described in these paper(s):

Jia, F.; Shi, C.; He, K.; Wang, C.; Xiao, B. Document Image Binarization Using Structural Symmetry of Strokes. In Proceedings of the 15th International Conference on Frontiers in Handwriting Recognition (ICFHR), Shenzhen, China, 23–26 October 2016; pp. 411–416.

Jia, F.; Shi, C.; He, K.; Wang, C.; Xiao, B. Degraded document image binarization using structural symmetry of strokes. Pattern Recognit. 2018, 74, 225–240.

I found only one Matlab-based implementation here: https://github.com/tanmayGIT/Bin_Structural_Symmetry_Strokes. But running this leads to errors and comments in the code indicate that the author was unsure about how to implement certain parts of the original papers.

So this is my question: are there any working implementations or tools available that implement the SSP method? (Programming language or platform don't matter as nowadays you can get nearly anything running on any platform.)

So far I was not able to find any implementation. Any hints where to find one are appreciated.

$\endgroup$

1 Answer 1

0
$\begingroup$

There is a precompiled Windows executable available on Github by one of the authors of the SSP paper: https://github.com/FuxiJia/DocumentBinarizationSSP/

This seems to be the only available implementation of the SSP algorithm (besides the Matlab approach).

I contacted the authors of "Robust Combined Binarization Method of Non-Uniformly Illuminated Document Images for Alphanumerical Character Recognition" and got the confirmation that this is what they used as well.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.