Institutional Repository of Vilnius University
Veido segmentavimo metodų efektyvumas /
Abstract
dc:descriptionThere is abundant literature suggesting various image segmentation systems, however, it seldom assesses the individual impact of the applied segmentation methods on the problem being solved. This study aims to compare different segmentation methods addressing the task of semantic segmentation of facial parts and to determine their accuracy, setting the encoder-decoder model as the benchmark. In this paper, the task of segmentation is reviewed, including the most significant achievements in solving this task, and the ongoing challenges related to model architecture, datasets, and human faces. Subsequently, the methods that could help to solve these specific problems inherent to the semantic segmentation of facial parts are analyzed. Finally, a reference model is defined, according to which segmentation methods are individually tested, and their accuracy is evaluated when performing semantic segmentation of facial parts. The research revealed that various methods, touching both the architecture of the encoder-decoder model and the modification of face training data, can impact the final segmentation quality. However, the application of individual methods brings minimal benefits in solving the problem of semantic segmentation of facial parts.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bukauskas, Antanas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192845239&prefLang=en_US