Document Image Dewarping and AutoCrop

Status: open
Supervisors: Florian Kleber, Markus Diem and Stefan Fiel

Problem Statement

Current Document Image Analysis (DIA) frameworks have also to deal with photographed images, which can originate from DSLR cameras or even from smartphones. Typically, the images are not taken within a defined aquisition setup and thus can have perspective distortions and an uneven illumination. The following picture shows a typical document image,



The goal of the Praktikum/Bachelor thesis is the detection of the document page, and the rectification as well as the cropping. The final document image should contain no background information and the perspective distortion should be corrected.


  • Literature research
  • Implementation in Matlab or C++ with OpenCV
  • Development and evaluation of the system
  • Written report or bachelor thesis (in English) and final presentation


  • Matlab, or C++ and OpenCV