ScanEraser: Lightweight Neural Networks for On-device Handwriting Erasure

The ScanEraser architecture consists of two fundamental components: a macrofield cleaner and a microfield refiner. The macrofield cleaner implements a mask-based handwriting removal mechanism with explicit erasure guidance. A feature-grabbing block enhances the capture of contextual features around handwritten text. The network is compact and efficient, trained using a Generative Adversarial Network (GAN). The microfield refiner employs a progressive refinement network with constrained receptive fields to repair detailed textures that the macrofield cleaner cannot address adequately.

Data preparation

The Epaper and Ebook datasets used for handwritten text removal research can be downloaded through the following links:

Epaper - Baidu Cloud (Password : a653 )

Ebook - Baidu Cloud (Password : a653 )

Note: This datasets can only be used for non-commercial research purpose. The training sets are available now, but requires a password to unzip. To use the databases, please fill out the agreement form and send it via email to us(666yyw666@gmail.com, 2371416@stu.neu.edu.cn). We will provide the unzipping password after receiving and approving your request.

Training

Once the data is well prepared, you can begin training:

python train.py

Testing

If you want to predict the results, run:

python test.py

Citation

If you find our method or dataset useful for your reserach, please cite: