Pasivna detekcija živih lica za sustave prepoznavanja lica

master's thesis

• Title (english): Passive Liveness Detection for Face Recognition Systems
• Author: Petra Omrčen
• Mentor: Igor Sunday Pandžić (mentor)
• Committee member: Igor Sunday Pandžić (predsjednik povjerenstva)
• Committee member: Mirko Sužnjević (član povjerenstva)
• Committee member: Krešimir Matković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-02-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Sustavi raspoznavanja lica sve se više koriste za identifikaciju i razne kontrole pristupa. Ukoliko se ne koriste dodatne mjere sigurnosti, sustave za raspoznavanje lica bilo bi lako zavarati na način da se ispred kamere postavi slika ili video snimka te osobe. Cilj ovog rada je istraživanje i implementiranje potencijalnih metoda koje rješavaju ovaj problem. Odabrane su tri metode. Dvije se baziraju na analizi boje i teksture slike. Isprobani su različiti prostori boja (RGB, HSV i YCrCB) i deskriptor LBP za izvlačenje korisnih značajki iz slika. Jedna metoda koristi histograme i SVM za klasifikaciju, a druga konvolucijsku neuronsku mrežu koja se ne oslanja samo na ručno izvučene značajke te koristi manje dijelove slike u procesu učenja. Posljednja metoda koristi dubinu slike kako bi odredila radi li se o stvarnoj ili lažnoj slici osobe.

Abstract (english)

Facial recognition systems are increasingly used for identification and various access controls. If additional security measures are not used, it would be easy to fool facial recognition systems by placing a picture or video of that person in front of the camera. The aim of this work is to research and implement potential methods that solve this problem. Three methods were chosen. Two are based on the analysis of the color and texture of the image. Different color spaces (RGB, HSV and YCrCB) and LBP descriptor were used to extract useful features from the image. One method uses histograms and SVM for classification, and the other a convolutional neural network that does not rely only on manually extracted features and uses patches of the image in the learning process. The last method uses the depth of the image to determine whether it is a real or spoof image.

• Keywords: Računalni vid; Konvolucijske neuronske mreže; Strojno učenje; Stroj potpornih vektora; RGB; HSV; YCrCb; Lokalni binarni uzorci; slike lica
• Keywords (english): Computer vision; Convolutional neural networks; Machine learning; Support vector machine; RGB; HSV; YCrCb; Local binary pattern; Face images
• Language: croatian
• URN:NBN: urn:nbn:hr:168:010137
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-07-13 10:25:12