University of Twente Student Theses

Login

Facial landmark detection under challenging conditions

Meijerink, Carlijn (2021) Facial landmark detection under challenging conditions.

[img] PDF
1MB
Abstract:In the facial identification process, for example, when examining evidence in the court of law, human experts are still used. This is a time-consuming process and therefore this study focuses on the possibility of using dlib, a facial landmark detector, for this. A landmark detector indicates key landmarks on the face and can be used to localize important facial regions. A comparison will be made between dlib and expert annotations on a variety of photos. This study focuses specifically on the influence of performance by the following conditions that can decrease the clarity of a face; illumination, resolution, quality, pose of the head, and color. Furthermore, three FISWG characteristic descriptors that can be abstracted by these landmarks; the eyebrow shape similarity, the intercanthal distance, and the left palpebral fissure, are tested for accuracy compared to the dlib annotations on a clear frontal image. The results of this study indicate that the different conditions influence the error rate by a human expert very little. The dlib error rate is influenced, mainly by very low resolutions and turned head poses. Dlib does show better error rates than an expert at the higher resolutions. For the FISWG characteristic descriptors, the challenging conditions shown very little influence on the accuracy.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/86867
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page