Knowledge Discovery Group

Novel Text Extraction from Infographics using Neural Networks



Master's thesis




Infographics – like other images – are indexed via their surrounding text. However, infographics often contain information that is not contained in the surrounding text. Part of the information contained in an infographic is in the form of figure text. Therefore, it is reasonable to improve the understanding of infographics by extracting text elements from the graphics.

Existing research on analyzing information graphics assume to have a perfect text detection and extraction available. However, text extraction from information graphics is far from solved. Classic linear approaches still lack enough extraction quality. The application of neural networks showed promising results for many computer vision task. Thus, a neural network based approach shall be tested to detect and extract multi-oriented text from infographics. We provide an annotated dataset of 440 infographics, which is extracted from an open access corpus of scientific publications and other sources. The results will be compared with a classic approach that we developed.


In this thesis, you will develop one or multiple neural networks to detect and extract text from infographics. This includes the definition and preparation of the input to the neural network as well as the structure itself. A thorough evaluation and comparison with classic approaches will conclude the work. We will provide the required hardware to efficiently train the model and run the evaluation.


In more detail, the work should cover:

  • Development of neural network(s) for text detection/extraction from infographics

  • Evaluation of the developed model and comparison with classic approaches




  • Good programming skills

  • Knowledge of machine learning techniques is an advantage (neural networks)

  • Knowledge of computer vision methods is beneficial



Got Interested? Send an email!


  • Homepage kicked off!