Hey all,
So over the past month I've really upped my scanning game (bought an IX1500 on MigJmz's suggestion and also after lurking the forums better). I'll be posting a thread listing my scans over in the Magazines section soon (lotsa of U.S. PSM!)
I've been trying to be careful about scanlines, bent corners, and other imperfections. But it is time consuming to check each page. I'm up for the effort, but it got me thinking: has anyone tried using any object detection programs to detect defects in scanned images?
I'm going to be dabbling in training a Tensorflow classifier (link to video example) for these things. I think that could really help scanners quickly note which pages they need to set aside for re-scan. So two asks:
- If you have tried this, or know anyone who has, please send advice or point me to those resources so I can contribute! (I very well could have missed the thread on this because I'm bad at searching)
- If this hasn't been tried before, please send me all your scanned images with imperfections! Training classifiers requires a lot of images, and finding images of imperfections online hasn't been very fruitful—especially scanlines/vertical streaks. And if you have the time, create a copy of the image cropped down just to the imperfection and send along both. Otherwise I'll be doing a lot of cropping.
I'm not sure how successful this will be, especially since I've never tried anything with Tensorflow before. But I think this could be a fun project that may also help the community. Let me know if you have ideas, thoughts, questions, or images to send over!