What is image dithering and how it is a problem to OCR

Recommended scanner settings for OCR of common office documents are 300dpi resolution, grayscale color mode (256 or more colors). Sometimes the user may reduce the number of colors for reducing scan file size. For 2-color documents (black and white), 16-color documents, etc. the scanner software might offer a "dithering" option that makes the image look better on screen for the human eye.

Figure 1.

Original image (grayscale, 256 colors)

1.png

Figure 2.

Dithered image (black-and-white, 2 colors)

dither_it_1.png

Let us take a closer look at the second image

Figure 3.

Dithered image enlarged.
2.png

You can see that the image barely makes sense. This is how OCR sees it. This explains why the dithered image is a problem for further recognition.

How to roll back to original image quality?

You can see that the image barely makes sense. This is how OCR sees it. This explains why the dithered image is a problem for further recognition.

Figure 4.

"Gaussian blur" applied to the enlarged image.

Gaussian_blur.png

The text became more readable, but the reading confidence will be far from perfect anyway.

What are the best scanning settings? 

Always use the "level"/"threshold" algorithm for black-and-white image scanning, and your result will be suitable for OCR.

Figure 5.

The same image with "level" dithering.

level_dithering.png

This version will be OK for OCR.

Was this article helpful?

0 out of 0 found this helpful

Have more questions? Submit a request

Comments

0 comments

Article is closed for comments.

Recently viewed