Urejanje zbirke fotografij z razpoznavo obrazov

Gaube, Veno (author), Čehovin Zajc, Luka (mentor)

Diplomsko delo se ukvarja z urejanjem osebnih arhivov fotografij, ki postajajo preveliki, da bi jih lahko v celoti uredili na roke. Razvit sistem zazna obraze ter jih preslika v opisni prostor opisnih vloženk z uporabo globoke konvolucijske nevronske mreže. Nato z gručenjem iz vloženk določimo identitete, ki jih lahko uporabnik v nadaljevanju ureja. Sistem smo ovrednotili na podlagi dveh zbirk slik, javni in zasebni, analizirali smo dve metodi gručenja pri naboru parametrov. Rezultati predstavljajo osnovo za nadaljnjo delo na sistemu za avtomatsko urejanje zbirk fotografij.

• Language: Slovenian
• Keywords: FaceNet, Python, Chinese Whispers, HDBSCAN, DBSCAN, vložitev, računalniški vid, prepoznava obraza, klasifikacija obraza, nevronske mreže, zbirka slik, urejanje
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2020
• PID: 20.500.12556/RUL-120055
• COBISS.SI-ID: 31257347
• Publication date in RUL: 15.09.2020

Secondary language

• Language: English
• Title: Organizing photo collection with face recognition

Abstract

The thesis deals with editing personal photo archives that are becoming too large to be completely edited by hand. The developed system detects faces and maps them into the space of descriptive embeddings using a deep convolutional neural network. Then, by clustering the embeddings, we determine the identities that the user can edit later. The system was evaluated on two collections of images, one public and one private, we analyzed two clustering methods with the set of parameters. The results represent the basis for further work on the system for automatic organization of photo collections.

• Keywords: FaceNet, Python, Chinese Whispers, HDBSCAN, DBSCAN, embedding, computer vision, face recognition, face classification, neural networks, image collection, ordering