[Python] CV2 image thresholding and filtering

I am making a simple OCR and I want to have better Thresholding. I have these scans of an old book and I want to be able to extract the letters as cleanly as possible. Unfortunately, all the thresholding and filtering techniques I've used still yield in significant noise.

The techniques I've used are the following:

1. mean Threshold + meadian Filtering
2. mean Threshold + morphological Closing
3. mean Threshold + Bilateral Filtering
4. mean Threshold + Non-Local Means Denoising
5. Gaussian Threshold + meadian Filtering
6. Gaussian Threshold + morphological Closing
7. Gaussian Threshold + Bilateral Filtering
8. Gaussian Threshold + Non-Local Means Denoising
9. otsu

Here is the original image:

Original Image

What are some techniques that could work way better on this?

Continue reading...