Batch scanning and correcting


#1

So I collect cabinet cards. If you are unaware of what these are they are a historical kind of photograph. I have a link to some sample scans if you want to take a look:

https://anansitech.com/Posting2.zip

I’ve been working on cataloging them and scanning them for a while now. I’ve been playing around with Retrobatch Pro and it’s awesome, but I can’t figure out how to do a few things I would love to do (if it’s possible).

  1. It would be great to have it be able to use the machine learning to decide if the image needed to be horizontal or vertical based on the figures in the image. But I am not sure if this is possible, and the site https://coreml.store linked from this page: https://flyingmeat.com/retrobatch/ seems to be dead.

  2. It would be cool it it could then examine the top and get it as parallel as possible (some images are slightly skewed, but aren’t always 100% straight edge regardless).

  3. Turn the extra area around the card to an alpha channel (there would have to be a variable amount of sensitivity and of distance).

  4. Trim to the edges.

  5. Output three sizes of the scans (renaming them sequentially).

  6. Move originals to a “done” folder.

I can do a lot of this manually, but it takes me 1-10 minutes per scan and I have a thousand cards. I included the workflow I created, so you can see what I am trying to do (mostly).

Any guidance would be appreciated.


#2

It’s possible, but I don’t have a model for you to use. I hope to add something to Retrobatch to train your own models in the near future.

I’m bummed that cormel.store is down! It had a bunch of good models to use.

The extra area you’re referring to- that would be generated from the un-skewing of the image?

This is already in RB. Have you tried it out?

This is possible, you just need to have multiple resizes coming off of a node (turning on manual connections in preferences will help do this).

I’ll have to think on how to do this one.


#3

Looks like some of the models have shown up here: