Vpliv poravnave na uspešnost razpoznavanja uhljev

RIBIČ, METOD (author), Peer, Peter (mentor), Štruc, Vitomir (comentor)

Uhlji kot biometrična modalnost so postali pomemben vir za samodejno razpoznavo oseb, predvsem v scenarijih nadzornih aplikacij, kjer se obraz ne vidi frontalno. V preteklih letih je priljubljenost metod lokalnih deskriptorjev narasla zaradi njihove invariantnosti na osvetlitev in zakrivanje, vendar pa te metode za vhodne podatke zahtevajo poravnane in vnaprej obdelane slike, kar pa lahko predstavlja velik problem zaradi kota, pod katerim so bile slike zajete. V tem diplomskem delu smo testirali, kako poravnava uhljev z metodo soglasja naključnega vzorca (RANSAC) in metodo kaskadne pozicijske regresije (CPR) vpliva na razpoznavo oseb na podlagi uhljev. Poravnava se je izvajala na podatkovni bazi uhljev AWE. Dokazali smo, da poravnava uhljev pozitivno vpliva na razpoznavo samo v primeru, ko so obravnavane slike zajete pod majhnim nagibom in odklonom. Slike, zajete pod večjim nagibom in odklonom, pa so po naših ugotovitvah prezahtevne za poravnavo in poslabšajo rezultate razpoznave, zaradi česar bi bilo potrebno uporabiti drugačne pristope za poravnavo.

• Language: Slovenian
• Keywords: računalniški vid, biometrija, razpoznavanje, poravnava, uhlji
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2016
• PID: 20.500.12556/RUL-91248
• Publication date in RUL: 27.03.2017

Secondary language

• Language: English
• Title: Influence of alignment on ear recognition

Abstract

Ear as a biometric modality presents a viable source for automatic human recognition especially in surveillance scenarios where face is not seen frontally. In recent years local description methods have been gaining on popularity due to their invariance to illumination and occlusion. However, these methods require that images are well aligned and preprocessed as good as possible. This causes one of the greatest challenges of ear recognition: sensitivity to pose variations. In this paper we test the influence of alignment on recognition performance on images from recently presented Annotated Web Ears dataset with alignment methods Random sample consensus (RANSAC) and Cascaded Pose Regression (CPR). We prove that alignment improves recognition rate but only on images with small angle on roll and yaw axis. On other pictures RANSAC and CPR fails to align ears and recognition rate is therefore lower versus unaligned pictures. Those pictures should be addressed with more advanced alignment methods in order to improve recognition rate.

• Keywords: computer vision, biometry, recognition, alignment, ears