Could someone point out the core texts or articles most useful on techniques for the removal of noise from scanned text for OCR applications?
-
4$\begingroup$ While this is out of my area of expertise, that sounds like a very broad question. What sort of noise are you trying to remove? Dust present on the paper during the scan? Noise from the image sensor? Something else? Grayscale? Color? Monochrome? $\endgroup$– Jason RDec 3, 2011 at 1:42
-
$\begingroup$ They are black and white scans of old books with discolored pages and the pages end up with black small random shaped spots. $\endgroup$– Harry SpierDec 4, 2011 at 0:32
1 Answer
Google research has some excellent papers, see for example:
An Overview of the Tesseract OCR Engine
Also, it seems that stackoverflow has a similar question:
https://stackoverflow.com/questions/4180629/ocr-and-image-preprocessing-techniques
The most powerful image filtering techniques that I'm aware are graph-cuts based, which run in the following steps:
- Calculate a (sparse) distance matrix between pixels (based on their intensity)
- Spectral Clustering, keep only the lowest 3-8 eigenvectors
- K-means clustering of the eigenvectors