Tenzorske metode u prepoznavanju lica

master's thesis

• Author: Suraja Poštić
• Mentor: Zlatko Drmač (mentor)
• Committee member: Zlatko Drmač (predsjednik povjerenstva)
• Committee member: Zoran Vondraček (član povjerenstva)
• Committee member: Slaven Kožić (član povjerenstva)
• Committee member: Sonja Štimac (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2018-09-24, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics

Abstract

Rad opisuje algoritam za prepoznavanje lica temeljen na tenzorima. Višedimenzionalna algebra omogućila nam je da uzmemo u obzir raznolike uvjete u kojima osoba može biti uslikana. Razlikovali smo izraze lica, osvjetljenja i kuteve gledišta. Prepoznavanje nepoznatog izraza lica obavljeno je vrlo uspješno. Suočavanjem algoritma s drugačijim osvjetljenjem ili kutem gledišta točnost klasifikacije značajno se smanjila. Redukcijom dimenzije osvjetljenja ublažili smo izražene svjetlosne kontraste. Posjedujemo li bazu slika s velikim brojem različitih osvjetljenja ovaj postupak možemo provesti prije same klasifikacije. Time izbjegavamo nebitne razlike među slikama iste osobe. Isto vrijedi i za ostale dimenzije u kojima je aproksimacija nižim rangom logična.

Abstract (english)

This thesis describes tensor based algorithm for face recognition. Multilinear algebra enabled us to take into consideration various scene conditions usually found in images. We distinguish face expressions, illumination and view points. Recognition of unfamiliar face expression was very successful. Facing with different illumination or viewpoint resulted in a significant accuracy drop. Reducing the dimension related to illumination we have managed to smooth noticeable light contrasts. In case we posses image base with many different illumination conditions we can conduct this procedure prior to the classification process. As a result we avoid inessential differences between images of the same person. Same applies to another dimension where dimensionality reduction makes sense.

• Keywords: algoritam za prepoznavanje lica; tenzor; višedimenzionalna algebra
• Keywords (english): face recognition algorithm; tensor; multilinear algebra
• Language: croatian
• URN:NBN: urn:nbn:hr:217:774106
• Study programme: Title: Finance and Business Mathematics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-06-12 11:19:05