Topic of your interest

Project describing graphic of DeepQuality

Machine Learning Framework for dynamic image style evaluation

Short name: DeepQuality

Fogra no. 13.004
Project leader: M. Wimmer (Fogra) and Prof. Dr D. Merhof (LfB)
Partner: Institute of Imaging & Computer Vision at Aachen University
Funding: BMWK (IGF) via AiF

Timescale: 01.06.2020 - 31.05.2022

Research Report

Objectives and Relevance

This research project had a closer look at the possibilities of automatization for professional retouch processes. These were investigated with the help of machine learning and artificial intelligence (AI). Currently available tools on the market (2022) offer already a wide range of possibilities to improve the retouching process with AI but often lack in quality. Especially for professional retouches quality is often too low. The main reason for this lies in the data used for training the AI system. This is exactly where professional retouch companies can take a profit. They often have their big, own, and qualitatively high image material, which can be used to train AI systems. This research project shows, how companies can exploit their data resources and implement their own AI systems. That will help to automatize a lot of processes without losing any kind of quality.

Solution Steps

This research project shows how AI may improve, simplify and speed up the professional retouch process. The resulting quality of used AI systems heavily depends on the data used for training the AI system. If the image quality of the used data is low, also the results will be quite unsatisfying. On contrary, if image quality is quite good, one can also expect quite good results from AI systems. In the best case, the data is given in form of image pairs. One image shows the content before retouch and the other one afterward. In some cases, it may be possible to use data from two different categories, but not necessarily in pairs. In this research project, only data-driven AI systems are explored. This means that the quality of learned transformations depends heavily on data used for training.

Image of project leader Martin Wimmer

Martin Wimmer

Prepress Technology

+49 89 431 82 - 411

Get in touch

Your contact

I would be pleased to answer your questions about the project and look forward to exchanging ideas with you.


The results of this research project were presented in an extensive report.

Further, the developed programming code is made public available in such a way that everyone can access, use and edit it easily directly in the web browser.

Meeting documents

TitleVersionDateFile typeDownload
Slides of PA-meeting 2021-03-01 V.1 01.03.2021 pdf Download
Slides of PA-meeting 2022-01-18 19.01.2022 pdf Download