University of Twente Student Theses

Login

Exploration of autoencoder as feature extractor for face recognition system

Bhaswara, I.D. (2020) Exploration of autoencoder as feature extractor for face recognition system.

[img] PDF
7MB
Abstract:Face recognition has been a challenging research problem due to many variations, for example occlusions, illuminations, poses, and expressions. In this paper, we review one of the unsupervised learning methods called autoencoder to be used as feature extractor for face recognition system. We explore several types of autoencoder, including regular and generative model, and take quantitative measurements on reconstruction and recognition of face images. Experimental results on Face Recognition Grand Challenge dataset show that there is a potential ability in using autoencoder as feature extractor for face recognition. Furthermore, apart from the latent variable dimensions, the encoder and decoder network of the autoencoder have an important role in the reconstruction and recognition performance. We also found that generative autoencoder model gives better clustering against identity of a subject. In addition, we apply residual network in the generative autoencoder model. We called this Resnet-WAE. It performs better in reconstruction and recognition and achieves area under the curve score of 0.8763 using likelihood ratio classifier. In the end, Resnet-WAE demonstrates promising results of using generative model as feature extractor in face recognition system.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Electrical Engineering MSc (60353)
Link to this item:https://purl.utwente.nl/essays/83138
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page