Prof. Dr. Tobias Friedrich

Deep Learning for Optical Character Recognition

MSc Project Seminar - Winter 2022/23


Optical Character Recognition (OCR) is a task in computer vision where one is given an image of a word and the task is to recognize the letters of the word. Recent advancement in deep learning improved significantly the state of the art in OCR, combining convolutional neural networks, recurrent neural networks and vision transformers. A specific case of OCR is when the word is handwritten, in that case the task is called Handwritten Text Recognition (HTR). For example, many historical documents are handwritten and are difficult to recognize. In this course we will learn and research about OCR and HTR.


There are no formal requirements. However, we expect participants to be familiar and enjoy working with deep learning. Please be aware that this course is taught in English. Furthermore, please note that there is a limitation to the amount of participants.


The evaluation of the course is fully based on the project.


Note that the course will be held fully virtually. The first meeting will be on Oct. 26th at 9:15 (please note the changed date). Here, we will discuss the projects and further meetings. The technical details for the meetings will be shared via moodle.

Lecture Team

The following persons are involved in this lecture:

Dr. Sarel Cohen


Office: K-2.18
E-Mail: Sarel.Cohen(at)hpi.de

Teaching Assistant

Office: K-2.09/10
E-Mail: Vanja.Doskoc(at)hpi.de