Deep Halftoning with Reversible Binary Pattern

ICCV Paper | Project Website | BibTex

Overview

Existing halftoning algorithms usually drop colors and fine details when dithering color images with binary dot patterns, which makes it extremely difficult to recover the original information. To dispense the recovery trouble in future, we propose a novel halftoning technique that dithers a color image into binary halftone with decent restorability to the original input. The key idea is to implicitly embed those previously dropped information into the binary dot patterns. So, the halftone pattern not only serves to reproduce the image tone, maintain the blue-noise randomness, but also represents the color information and fine details.

Run

1. Requirements:

   • Basic variant infomation: Python 3.7 and Pytorch 1.0.1.
   • Create a virutal environment with satisfied requirements:

     conda env create -f requirement.yaml

2. Training:

   • Place your training set/validation set under dataset/ per the exampled file organization.
   • Warm-up stage (optional): Pre-download the checkpoint of feature extractor Here.

     python train_warm.py --config scripts/invhalf_warm.json

     If this stage skipped, please download the pretrained warm-up weight and place it in checkpoints/, which is required at joint-train stage.

   • Joint-train stage:

     python train.py --config scripts/invhalf_full.json

3. Testing:

   • Download the pretrained weight below and put it under checkpoints/.
   • Place your images in any accesible directory, e.g. test_imgs/.
   • [Halftoning]: Dither the input images into binary halftones

     python inference.py --model checkpoints/model_best.pth.tar --data_dir ./test_imgs --save_dir ./result

   • [Restoration]: Restore the generated halftone back to RGB images

     python inference.py --model checkpoints/model_best.pth.tar --data_dir ./test_imgs --save_dir ./result --decoding

Copyright and License

You are granted with the LICENSE for both academic and commercial usages.

Citation

If any part of our paper and code is helpful to your work, please generously cite with:

@inproceedings{xia-2021-inverthalf,
    author   = {Menghan Xia and Wenbo Hu and Xueting Liu and Tien-Tsin Wong},
    title    = {Deep Halftoning with Reversible Binary Pattern},
    booktitle = {{IEEE/CVF} International Conference on Computer Vision (ICCV)},
    year = {2021}
}