Klasifikacija slika dubokim modelima i Fisherovim vektorima

master's thesis

• Title (english): Image classification with deep models and Fisher vectors
• Author: Ivana Viduka
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Jan Šnajder (član povjerenstva)
• Committee member: Slobodan Ribarić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2017-07-17, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Za ovaj rad su opisane i implementirane dvije metode klasifikacije slika. Prvu metodu čini duboki model temeljen na arhitekturi VGG16 konvolucijske mreže čiji su parametri inicijalizirani težinama naučenima na slikama ImageNet baze. Drugi pristup uključuje konvolucijske značajke dobivene uz pomoć javne parametrizacije VGG19 mreže iz kojih se generiraju Fisherovi vektori, a zatim se provodi klasifikacija pllitkim klasifikatorom. Usporedbom rezultata klasifikacije dobivenih učenjem i testiranjem na dva skupa slika koji prikazuju scene s hrvatskih prometnica - Zebre i Znakovi, zaključuje se da end-to-end pristup učenju klasifikatora kakav je prisutan kod konvolucijske mreže donosi puno bolje rezultate.

Abstract (english)

Two different image classification methods are described and implemented for this work. The first method is a deep model based on the architecture of VGG16 convolutional network, having parameters initialized with weights learned on images from ImageNet database. The second approach includes Fisher vectors generated from convolutional features that are extracted with the help of the publicly available pretrained VGG19 network. Shallow classifier is then used for the classification of Fisher vectors. Comparing the classification results obtained for images from two datasets containing traffic scenes of Croatian roads, we draw a conclusion that the end-to-end approach to the image classification, like the one used in our deep model, shows significantly better results.

• Keywords: klasifikacija slika; konvolucijske neuronske mreže; jezgrene funkcije; Fisherovi vektori; linearni klasifikatori
• Keywords (english): image classification; convolutional neural network; kernel functions; Fisher vectors; linear classifiers
• Language: croatian
• URN:NBN: urn:nbn:hr:168:084270
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-14 01:56:09