magistrsko delo
Mark Lukek (Author), Vitomir Štruc (Mentor), Klemen Grm (Co-mentor)

Abstract

V zadnjem času so konvolucijski modeli, ki temeljijo na nevronskih mrežah, do- segli velik uspeh pri superločljivosti z uporabo ene vhodne slike (ang. Single- Image-Super-Resolution). Takšni modeli so zelo prožni in učinkoviti pri neline- arni preslikavi slik nizke ločljivosti v slike visoke ločljivosti. V tem delu pred- stavljamo nov postopek za superločljivost, ki temelji na dveh samokodirnikih in sklopljenih latentnih prostorih. Prvi samokodirnik je zmožen rekonstrukcije nizkoločljivostnih slik, drugi pa visokoločljivostnih slik. Latentna prostora samo- kodirnikov povezuje povezovalna mreža, ki omogoča pretvorbo med nizko- in viso- koločljivostnim latentnim prostorom. Z uporabo nizkoločljivostnega samokodir- nika, povezovalne mreže in visokoločljivostnega samokodirnika je mogoče poljubno vhodno nizkoločljivostno sliko preslikati v sliko visoke ločljivosti. Rezultati ome- njene metode so testirani na štirih podatkovnih zbirkah, CASIA-WebFace, LFW, QMUL-TinyFace in QMUL-SurvFace. Del podatkovne zbirke CASIA-WebFace je bil uporabljen za učenje vseh modelov, preostali del za testiranje. Zbirki QMUL- TinyFace in QMUL-SurvFace sta uporabljeni za preverjanje delovanja sistema na realnih slikah, kjer nimamo visokoločljivostnih parov. Rezultati izvedbe su- perločljivosti se na koncu še primerjajo z že obstoječimi pristopi, kot so bikubična interpolacija, SRCNN in SRGAN. V primerih sprednjega dela obraza, naš pri- stop presega delovanje bikubične interpolacije in modela SRCNN. Obrazi so bolj izraziti in gladki, vendar ne vsebujejo dovolj visokoločljivostnih podrobnosti, kot jih proizvede sistem SRGAN.

Keywords

globoko učenje;umetna inteligenca;konvolucijski sloj;superločljivost;samokodirnik;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FE - Faculty of Electrical Engineering
Publisher: [M. Lukek]
UDC: 004.8(043.3)
COBISS: 105855747 Link will open in a new window
Views: 150
Downloads: 33
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Improving the resolution of facial images using latent-space coupled autoencoders
Secondary abstract: Recently, convolutional models based on neural networks have achieved great success in super-resolution using a single input image, called Single-Image-Super- Resolution or SISR. Such models are very flexible and efficient in non-linear map- ping of low-resolution images to high-resolution ones. In this work, we present a novel super-resolution procedure based on two autoencoders and coupled latent spaces. The first autoencoder is capable of reconstructiong low-resolution images, while the second one is capable od reconstructing high-resolution images. The latent spaces of the two autoencoders are connected by a linking network which allows conversion between the low- and high- resolution latent spaces. Using the low-resolution encoder, the linking network and the high-resolution decoder it is possible to efficiently upscale an arbitrary low-resolution inpout image. The re- sults of the above method area tested on four datasets, CASIA-WebFace, LFW, QMUL-TinyFace and QMUL-SurvFace. Part of the CASIA-WebFace database was used to train all models, the rest for testing. The QMUL-TinyFace and QMUL-SurvFace databases are used to verify the system performance on real images where we do not have high-resolution pairs. Finally, the results of the super-resolution model are further compared with existing approaches such as bicubic interploation, SRCNN and SRGAN. In the cases frontal face images are used as input, our approach outperforms bicubic interpolation and the SRCNN model. The faces are more pronounced and smoother, but do not contain less high-resolution details than faces produced by SRGAN.
Secondary keywords: deep learning;artificial intelligence;convolutional layers;super- resolution;autoencoders;
Type (COBISS): Master's thesis/paper
Study programme: 1000316
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za elektrotehniko
Pages: XVIII, 65 str.
ID: 15088746