20.500.12556/RUL-164243 Prepoznavanje globokih ponaredkov z enorazrednim učenjem Deepfake detection using one-class learning V okviru magistrskega dela smo si zadali zgraditi model, ki bi prepoznaval globoke ponaredke. Modeli za prepoznavanje globokih ponaredkov so v splošnem boljši, če so naučeni na čim več različnih zbirkah globokih ponaredkov, ki so ustvarjeni s čim več metodami za ustvarjanje le-teh, ker to pomeni, da bolje generalizirajo. Naš pristop za doseganje boljše generalizacije pa je enorazreden, kar pomeni, da za predstavitev problema uporabimo samo razred resničnih slik. Iz teh slik izdelamo sintetične slike ponaredkov s pomočjo metode Self-Blending Images. Za nevronske mreže velja, da niso enostavno interpretativne, zato smo v proces učenja dodali segmentacijo. Naš model označi del obraza na sliki, za katerega predvideva, da je bil ponarejen, in na podlagi pravilnosti te maske se model uči prepoznavati globoke ponaredke. Maske, ki jih naš model ustvari, ocenimo na šestih zbirkah. Klasifikacijo modela ocenimo na podatkovnih zbirkah, na katerih ocenjujejo tudi avtorji drugih enorazrednih metod in v povprečju dosežemo najboljše rezultate. As part of our Master's thesis, we set out to build a model that could detect deepfakes. In general, models for identifying deepfakes are better when they are trained on as many different datasets of deepfakes as possible that are generated by as many methods for generating them as possible because this means that they generalise better. However, our approach to achieving better generalisation is one-class, meaning we only use a class of real images to represent the problem. From these images, we create synthetic fake images using the Self-Blending Images method. Neural networks are generally not considered to be easily interpretable, so we added segmentation to the learning process. Our model labels the part of the face in the image that it assumes has been forged, and based on the correctness of this mask, the model learns to identify deepfakes. We evaluate the masks generated by our model on six datasets. We evaluate the classification of the model on the datasets on which the authors of other models also assess, and on average we obtain the best results. prepoznavanje globokih ponaredkov enorazredno učenje segmentacija deepfake detection one-class learning segmentation true false false Slovenski jezik Angleški jezik Magistrsko delo/naloga 2024-10-17 16:25:02 2024-10-17 16:25:11 2024-11-15 09:40:56 0000-00-00 00:00:00 2024 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 37086 215122947 Kronovsek_Andrej_-_Prepoznavanje_globokih_ponaredkov_z_enorazrednim_ucenjem.pdf Kronovsek_Andrej_-_Prepoznavanje_globokih_ponaredkov_z_enorazrednim_ucenjem.pdf 1 E06837EDABD832E9C4163AE51DB68B87 b49153dc25f096af08f90f6f8a3f73cf81e37dd590140125487a9ab3f6d0e7c3 5df78972-8c93-11ef-b232-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=194497 Fakulteta za računalništvo in informatiko 0 0 0