Logo TU Ilmenau


Prof. Heidi Krömker


Telefon +49 3677 69-2890

E-Mail senden


Themen für Abschluss- und Projektarbeiten


Fachgebiet(e): AVT

Thema Nr.: 2019/23, erstellt am: 23.05.2019

Subjective Evaluation of DNN based video enhancements

Considering that today there are several DNNs available that can be used to replace traditional video enhancements algorithms, it is so far not clear if such DNN enhancements will lead to a higher perceived video quality.
Starting with two general video processing parts, i.e., frame interpolation and resolution upscaling [1], it should be analyzed which DNN based methods are suitable and can be applied.
Beside DNN based enhancements also classical state of the art algorithms, e.g., lanczos upscaling algorithm or motion frame interpolation should be included to form a valid comparison set.
After developing tools for such enhancements methods, is is required to design and conduct a subjective video quality test.
The results of the test will be used to analyze if DNN based enhancements are able to outperform classical methods on a subjective level.
It is clear so far that the new methods will not run in real-time, however this is not required for this topic.

## Tasks
* build up tools for using DNNs to perform frame interpolation and upscaling
* design suitable subjective tests using different enhancement methods
* analyze results of tests

## Requirements
* python3, ffmpeg knowledge
* data analysis skills
* test designing

## Literature
* 1: Ledig et al: "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network";
* 2: Jiang et al: "Super SloMo: High Quality Estimation of Multiple Intermediate Frames for Video Interpolation";

verantwortlicher Hochschullehrer:

Alexander Raake


Steve Göring


Steve Göring

zurück zur Liste