Sparkling Light Publisher
Sparklinglight Transactions on Artificial
Intelligence and Quantum Computing
(STAIQC)
Website: https://sparklinglightpublisher.com/ ISSN (Online):2583-0732
E-mail address of authors:
a
bm783m@att.com,
b
absa00002@stud.uni-saarland.de
© 2022 STAIQC. All rights reserved.
Please cite this article as: Biswas Mishra and Abhishek Samanta. (2022). Quantum Transfer Learning for Deepafake
Detection. Sparklinglight Transactions on Artificial Intelligence and Quantum Computing (STAIQC), 2(1), 17-27. ISSN
(Online):2583-0732. Received Date: 2022/05/02, Reviewed Date: 2022/06/10, Published Date: 2022/06/19.
Quantum Transfer Learning Approach for Deepfake Detection
Bishwas Mishra
a#
, Abhishek Samanta
b
a
Senior Software Engineer, AT&T, 4183 Grouse Court #104, Mechanicsburg, PA 17050, USA
b
Saarland Informatics Campus, Saarland University, Germany
Abstract
Deepfake image manipulation has achieved great attention in the previous year’s owing to brings solemn challenges from the
public self-confidence. Forgery detection in face imaging has made considerable developments in detecting manipulated images.
However, there is still a need for an efficient deepfake detection approach in complex background environments. This paper
applies the state-of-the-art quantum transfer learning approach for classifying deepfake images from original face images. The
proposed model comprises classical pre-trained ResNet-18 and quantum neural network layers that provide efficient features
extraction to learn the different patterns of the deepfake face images. The proposed model is validated on a real-world deepfake
dataset created using commercial software. An accuracy of 96.1 % was obtained.
© 2022 STAIQC. All rights reserved.
Keywords: Deepfake, Forgery detection, Quantum Neural Networks.
1. Introduction
Rapidly increase in the development of digital image processing field, image modification without visual apparent
is becoming easier [1]. These days, seeing does no longer believe. Recently, generative adversarial networks-based
imaging forensics attained much attention. Several forensic approaches such as hand-crafted and deep features are
commonly utilized for deepfake image classification [3]. Face images are commonly utilized in biological imaging
modalities, which comprise intuitive personal information [4]. However, facial images having susceptibility can be
forged easily [5]. Recently, significant progress has been made for deepfake image creation using generative
networks and computer graphics approaches [6]. These methodologies have been utilized in different facial imaging
applications [7]. These approaches can be utilized intentionally for malicious activities [8]. Existing facial imaging
approaches might be divided into three steps such as manipulation of the identity, expression, and transfer of the
attribute [9]. In the manipulation of the identity process, deepfake images are generated of the entire imaginary
person replaced one face with another image of the face through deepfake and the Face Swap [10]. In the
manipulation of facial expression, deepfake images are generated with a different expression and also transferred the
face expression from the source image to the destination image [11]. Extensive research has been done on deepfake