National Repository of Dissertations in Serbia
    • English
    • Српски
    • Српски (Serbia)
  • English 
    • English
    • Serbian (Cyrilic)
    • Serbian (Latin)
  • Login
View Item 
  •   NaRDuS home
  • Универзитет у Београду
  • Електротехнички факултет
  • View Item
  •   NaRDuS home
  • Универзитет у Београду
  • Електротехнички факултет
  • View Item
JavaScript is disabled for your browser. Some features of this site may not work without it.

Niskodimenzionalni prostorno-teksturalni deskriptori multispektralnih slika

Low-dimensional spatial-textural descriptors of multispectral images.

Thumbnail
2016
Disertacija.pdf (7.674Mb)
IzvestajKomisije.pdf (1.471Mb)
Author
Avramović, Aleksej
Mentor
Reljin, Irini
Committee members
Bjelica, Milan
Babić, Zdenka
Popović, Miodrag
Risojević, Vladimir
Metadata
Show full item record
Abstract
Prepoznavanje vizuelnog i semantickog sadrzaja u slikama primjenom racunarskih programa ima sve veci znacaj u raznim granama privrede i industrije, medicini, vojnoj industriji, itd. Prepoznavanje sadrzaja u slikama se u vecini prakticnih aplikacija oslanja na metode obrade koje na osnovu numerickih vrijednosti na digitalnim slikama odreduju njihov sadrzaj. U mnogim slucajevima vazno je odrediti koliko je sadrzaj dvije slike slican, da li prikazuju isti objekat ili isti dogadaj. Sa druge strane, svjedoci smo da se razvojem moderne tehnologije nezaustavljivo povecava broj generisanih digitalnih slika. Savremeni klinicki centri opremljeni digitalnom radiologijom, dnevno generisu i do nekoliko desetaka hiljada novih snimaka. Manuelno opisivanje sadrzaja tako velikog broja slika predstavlja praktican problem. Takode, svakodnevno dobijamo veliku kolicinu podataka snimljenih tehnikama daljinske detekcije, pri cemu specicne aplikacije zahtjevaju brzo prepoznavanje sadrzaja takvih snimaka. Potr...eba za prepoznavanjem vizuelnog i semantickog sadrzaja u slikama dovela je do razvoja velikog broja pristupa za opisivanje tog sadrzaja na nacin pogodan za koriscenje u racunarskim sistemima. Cesto se slikama pridruzuju odgovarajuci deskriptori koji treba da opisu" sadrzaj u slikama. Ti deskriptori su vektori numeri ckih vrijednosti ili skup kljucnih rijeci, koji treba da budu odredeni tako da se pomocu njih mogu razlikovati slike razlicitog vizuelnog ili semantickog sadrzaja ili prepoznati slike slicnog sadrzaja. Posto ljudski vizuelni sistem ekasno koristi informacije o teksturi za prepoznavanje objekata, u prakticnim aplikacijama se cesto koriste deskriptori teksture. Razvoj tehnologije omogucio je upotrebu jeftinih multispektralnih kamera, pa se postavlja pitanje kako opisati sadrzaj slika sa vecim brojem spektralnih opsega. Jednostavno prosirivanje deskriptora i upotreba dodatnih podataka moze posluziti da se na odgovarajuci nacin opise sadrzaj multispektralnih slika, ali sa znacajnim povecanjem potrebnih memorijskih resursa i racunarske kompleksnosti. U ovoj disertaciji su predlozene su metode za izdvajanje niskodimenzionalnih deskriptora multispektralnih slika, pogodnih za automatsku klasikaciju slika. Takode, razmotreni su pristupi za ukljucivanje podataka o prostornom rasporedu lokalnih obiljezja na slikama u deskriptor, kako bi se povecala tacnost klasikacije. Na kraju, predlozena je nova metoda za izdvajanje niskodimenzionalnih prostorno-teksturalnih deskriptora za multispektralne slike...

Recognition of visual and semantic content on images using computer programs gained an importance in various elds of agriculture, industry, medicine, military industry etc. Most practical applications use certain methods based on the numerical value of the digital images to determine what is a content of those images. In many cases, it is important to determine a level of visual or semantic similarity between two dierent images, does two images showing the same object or maybe the same event. We are witnessing that development of modern technology cause unstoppable increase of the number of daily generated digital images. Modern clinical centers, equipped with digital radiology, generate up to tens of thousands of new images per day, so their manual annotation presents a practical problem. Moreover, each day brings a large amount of remotely sensed images and many specic applications require fast identication of their visual content. The need to recognize visual and semantic content in... images initiated the development of a large number of methods for description of image contents, in such a manner suitable for use in specic computer systems. Images are is associated with appropriate descriptors that should describe" the visual or semantic content of those images. These descriptors can be vectors with numerical values, which should be calculated so it is possible to use them to distinguish between images with dierent visual or semantic content or to recognize images with similar content. Since the human visual system eectively relies on texture to identify objects, texture descriptors are often used in practical applications. Technology development enable the widespread usage of cheap multispectral cameras, which can capture the data beyond visible spectra. Thus, it is necessary to investigate how to represent and describe the content of multispectral images in the way suitable for practical applications based on image classication. Simple extension of descriptors can increase classication accuracy, but with the cost of more memory resources and computational complexity. In this dissertation, dierent methods for extraction of low-dimensional descriptors for multispectral images are proposed, which used for automatic image classi cation. Moreover, the usage of spatial position of local textural features is discussed as well. It was concluded that extension of texture descriptor of grayscale images with additional data providing spatial-based texture features, can increase classication accuracy...

Faculty:
Универзитет у Београду, Електротехнички факултет
Date:
23-09-2016
Keywords:
Tekstura / Texture / descriptors / multispectral images / image classication / neural networks / object detection / deskriptori / multispektralne slike / klasifikacija slika / neuronske mreže / detekcija objekata
[ Google Scholar ]
Handle
https://hdl.handle.net/21.15107/rcub_nardus_6866
URI
https://nardus.mpn.gov.rs/handle/123456789/6866
http://eteze.bg.ac.rs/application/showtheses?thesesId=4021
https://fedorabg.bg.ac.rs/fedora/get/o:13646/bdef:Content/download
http://vbs.rs/scripts/cobiss?command=DISPLAY&base=70036&RID=48338191

DSpace software copyright © 2002-2015  DuraSpace
About NaRDus | Contact us

OpenAIRERCUBRODOSTEMPUS
 

 

Browse

All of DSpaceUniversities & FacultiesAuthorsMentorCommittee membersSubjectsThis CollectionAuthorsMentorCommittee membersSubjects

DSpace software copyright © 2002-2015  DuraSpace
About NaRDus | Contact us

OpenAIRERCUBRODOSTEMPUS