Supervisor: Fabian Hollaus

The quality of photographs of single document pages or book pages is often limited by distortions. Dewarping techniques can be used in order to remove such distortions, i.e. rectify the document images.

Goal

The aim of this work is to implement a document image dewarping algorithm. The method should make use of deep learning, whereby the overall aim is to generate a lightweight model that can be applied on mobile phones (via the DocScan app.

Workflow

Literature Review
Data Preparation
Implementation
Evaluation
Written Report/Thesis and final presentation

References

[1] Guo-Wang Xie, Fei Yin, Xu-Yao Zhang, Cheng-Lin Liu: Dewarping Document Image by Displacement Flow Estimation with Fully Convolutional Network. DAS 2020: 131-144