Razpoznavanje čustvenih izrazov s pomočjo globokih nevronskih mrež

diplomsko delo

Gregor Štefanič (Author), Božidar Potočnik (Mentor), Uroš Mlakar (Co-mentor)

Abstract

V diplomskem delu smo se ukvarjali z razpoznavanjem čustvenih izrazov z digitalnih slik obrazov. Razpoznavali smo med sedmimi čustvenimi izrazi, vključno z nevtralnim. Pregledali smo obstoječa dela na področju razpoznavanja čustvenih izrazov, preučili globoke nevronske mreže in pripravili arhitekturo, ki je primerna za razpoznavanje čustvenih izrazov s slik. Uporabili smo arhitekturo z residualno nevronsko mrežo. Našo rešitev smo razvili s pomočjo ogrodja TensorFlow in programskega vmesnika Keras. Implementirali in preizkusili smo jo na mešanih slikah iz podatkovnih baz JAFFE, CK in MMI. Natančnost razpoznavanja čustvenih izrazov na 1017 testnih slikah z našo nevronsko mrežo je bila v povprečju 99,3-odstotna, kar je primerljivo oziroma boljše od sorodnih del.

Keywords

razpoznavanje čustvenih izrazov;globoka nevronska mreža;računalniški vid;residualna nevronska mreža;diplomske naloge;

Data

Language:	Slovenian
Year of publishing:	2019
Typology:	2.11 - Undergraduate Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	[G. Stefanič]
UDC:	004.8:004.93(043.2)
COBISS:	22908694
Views:	603
Downloads:	117
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Facial expression recognition using deep neural networks
Secondary abstract:	In this thesis we dealt with facial expression recognition in digital facial images. We distinguished between seven different facial expressions, including neutral. We reviewed existing works dealing with facial expression recognition, examined deep neural networks, and prepared an architecture suitable for facial expression recognition. Our architecture uses a residual neural network. Our solution was developed using TensorFlow and Keras. We implemented and tested the network on mixed images from databases JAFFE, CK, and MMI. Accuracy of facial expression recognition in 1017 test images with our neural network was 99.3% on average, which is comparable or better than related works.
Secondary keywords:	facial expression recognition;deep neural network;computer vision;residual neural network;
Type (COBISS):	Bachelor thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages:	XII, 36 str.
ID:	11219731