Investigation of Deep Fake Images Using Pre-Trained CNN Frameworks

Introduction

Smart phones' rising complexity and the rise of social networks' digital object content have led to an enormous number of new digital object content in recent years. As a result of the widespread use of digital photographs, a proliferation of tools and methods for manipulating image data has occurred. Because these methods were tedious and time-consuming, they had previously been out of the reach of the majority of people. Artificial intelligence-based false picture production is now possible thanks to deep generative models, which allow for the rapid transmission of high-quality tampered media material (Li et al., 2020). In the field of image improvement and manipulation, one of the most promising recent developments has been the use of Generative Adversarial Networks (GANs). Splicing, resampling, and copy moving are all examples of digital image modifications that can be detected using digital image forensics tools and methodologies (Nataraj et al., 2019). Image augmentation and manipulation have been greatly improved by the use of generative adversarial networks (GANs). GANs are made up of two networks, one that generates false pictures and the other that assesses whether an image is real or fake (Chen et al., 2022). Digital picture forensics is devoted to identifying and preventing image forgeries in order to control the circulation of such fabricated material. ” According to various estimates, the internet receives over two billion photographs each day. Image forgery detection has made significant progress, but digital video falsification detection is still a challenging problem. Due to the significant deterioration of video frames when they are compressed, many image processing techniques cannot be applied to video. Digital image forensics has had success using deep learning. However, deep learning may be used to fake videos as well (Bonettini et al., 2020). There are inherent picture disparities across the borders that make face editing procedures fundamentally recognizable despite the standard process of blending a changed face into an existing backdrop (Durall et al., 2019). One of the most important issues in today's digital society is the manipulation of visual material. Face expression manipulation and identity manipulation are two of the most common ways of facial manipulation now in use. It is possible to communicate the facial emotions of one person to another in real time using only inexpensive equipment. This is the second type of facial forgery in which a person's face is replaced with an image of another person. However, DeepFakes uses deep learning to conduct face swapping as well. DeepFakes, on the other hand, must be trained for each pair of movies in order to do face swapping in real time. (de Lima et al., 2020). Massive advancements in the area of facial modification have shown a difficult footprint that may be exposed to the specifics of automated video editing techniques in the last few years. Forensic footprints are frequently quite delicate and difficult to detect, which has sparked a lot of interest in the procedures. New artistic possibilities are opened up by facial modification (e.g., movie creation, visual effects). However, criminal individuals may easily create video forgeries by manipulating the facial expressions. Detecting face modification using recent technologies is the focus of this article (Afchar, 2019).

This Chapter is organized as follows, background section gives a brief synopsis of the underlying research, main focus of the Chapter section discusses about the Deep learning frameworks that can be used to detect Deep fake images. solutions and recommendations section provide the results from the application of TENSOR FLOW and KERAS frameworks and the final section with future scope and conclusion.