Create New Bleed-Through Augmentation

[sparkfish]

Create New Bleed-Through Augmentation

[jboarman]

https://www.isos.dias.ie/Sigmedia/Bleed_Through_Database.html

Bleed-Through Database
The images contained here form a database that is designed to be a resource for people working in the field of digital document restoration, and more specifically on the problem of bleed-through degradation. It consists of a set of 25 registered recto verso sample grayscale image pairs, taken from larger manuscript images, with varied degrees of bleed-through.

The crops were taken such that they would contain a sentence or phrase of text, in order to make tests on improved legibility possible. All images in the database are saved in tif format. The verso side of each image pair has been flipped horizontally and registered to the recto side so that bleed-through text on one side and corresponding foreground text on the other are aligned.

One of the main problems encountered when researching digital document restoration is that no ground truth exists. That is there is no available clean original document with which to compare restoration results. Therefore ground truth has to be obtained either by creating synthetic degraded document images with known ground truth, or by creating synthetic ground truth images for real degraded images. This database contains the latter; manually created foreground text masks for each image have been created and are included here.

[kwcckw]

Yes, i think this is a good idea, i will try on this later.

[kwcckw]

Is something like this looks good enough?

[jboarman]

I think this sample looks great!

Can you provide some additional variety using other source images, and limit to B/W so that we can focus the evaluation on a more limited set of augmentations or effects?

[kwcckw]

Alright, i will open a pull request (#19) when it's done.