Download
PDF 1.85 MB
doctoral thesis
5   2
Analysis of the image moments sensitivity for the application in pattern recognition problems
Kristina Šariri (2010)
University of Zagreb
Faculty of Science
Department of Physics
Metadata
TitleAnaliza osjetljivosti momenata slike s ciljem primjene na problematiku raspoznavanja uzoraka
AuthorKristina Šariri
Mentor(s)Nazif Demoli
Abstract
Momenti slike su numerički deskriptori koji sadrže informaciju o svojstvima invarijantnim na translaciju, rotaciju, promjenu skale i neke oblike distorzije, a njihova analiza je jedna od metoda koje se često koriste pri analizi slika i raspoznavanju uzoraka. U okviru ove radnje razvijeni su algoritmi za računanje geometrijskih, Legendreovih, Zernikeovih, Fourier – Mellinovih te tri tipa Fourier – Jacobijevih momenata, kao i iz njih definiranih invarijanti slike u programskom jeziku MatLab uz rješavanje inverznog problema rekonstrukcije početnog ulaza. Za sve tipove momenata osim najjednostavnijih geometrijskih definirani su vektori osjetljivosti na rotaciju i promjenu skale čije su komponente oni članovi skupa koji nose značajnije informacije o ulaznoj slici. Primjenom novih deskriptora na klasifikaciju rukom pisanih slova i identifikacijskih fotografija osoba pokazano je da je relevantna informacija o ulazu na taj način sačuvana, a njihov je izračun znatno brži i jednostavniji uz zadržanu sposobnost jednoznačnog raspoznavanja uzoraka. Korištenjem momenata slike i vektora osjetljivosti analizirani su znakovi s dvaju glagoljskih spomenika te utvrđeno postojanje mješavine znakova trokutastog i okruglog modela glagoljice. Metoda je primijenjena i na klasifikaciju tragova puzanja ličinki mutanata vinske mušice za potrebe proučavanja odgovora živčanog sustava na različite podražaje.
Keywordsimage analysis pattern recognition invariant features image moments sensitivity vector
Parallel title (English)Analysis of the image moments sensitivity for the application in pattern recognition problems
Committee MembersDamir Veža (committee chairperson)
Goran Pichler (committee member)
Nazif Demoli (committee member)
GranterUniversity of Zagreb
Faculty of Science
Lower level organizational unitsDepartment of Physics
PlaceZagreb
StateCroatia
Scientific field, discipline, subdisciplineNATURAL SCIENCES
Physics
UDK53
NATURAL SCIENCES
Physics
Study programme typeuniversity
Study levelpostgraduate
Study programmePhysics
Academic title abbreviationdr. sc.
Genredoctoral thesis
Language Croatian
Defense date2010-11-24
Parallel abstract (English)
Image moments are numerical descriptors invariant to translation, rotation, change of scale and some types of image distortion and their analysis is one of the most often used methods in image processing and pattern recognition. In this work, algorithms for calculation of geometric, Legendre, Zernike, Fourier – Mellin and three types of Fourier – Jacobi moments were implemented in MatLab. Hu's, affine and blur invariants were also obtained as well as inverse problem of input image reconstruction solved. For each type of image moments exept geometric ones the set of sensitivity vectors for rotation and scale were defined. Their components are those image moments which describe more important features of the input image. These new descriptors were applied for classification of handwritten letters and identifying personal photos. It was shown that the process of such descriptor calculation is much faster and simpler while preserving all the relevant information about input image. Using this method, the signs carved in two glagolitic inscriptions were analyzed and the mixture of triangular and round glagolitic letters found. The method was also applied to classification of the mutant fruit fly larvae crawling trails which is needed in studying responses of the nervous system to different stimuli.
Parallel keywords (Croatian)analiza slike raspoznavanje uzoraka invarijantna svojstva momenti slike vektor osjetljivosti
Extent80 str.
Versionaccepted version
Resource typetext
Access conditionOpen access
Terms of usehttp://rightsstatements.org/vocab/InC/1.0/
Noteaccepted version
URN:NBNhttps://urn.nsk.hr/urn:nbn:hr:217:457438
CommitterIva Čizmin