DETECTION OF ATTRIBUTES MANIPULATION IN DEEP FAKE IMAGES USING HYBRIDLEARNING TECHNIQUE TO CLASSIFY INDIGENOUS AND FORGED IMAGES

Abstract

Deepfake technology employs artificial intelligence (AI) algorithms tocreatemanipulated photographs and videos that are indistinguishable fromauthenticones. Generative adversarial networks (GANs), a type of deep-learning algorithm, aredriving the development of deepfake, which have the potential to compromiseindividual privacy, as they can be used to create pornographic content bysuperimposing images.As a result, digital media, including news broadcasts, onlinevideo clips, and live streams, are experiencing trust issues. Recently, synthetic imagecreation and manipulation methods have improved, enabling the creation of realisticfake face images. Despite the emergence of certain deep learning-basedimageforensic techniques, it is still challenging to differentiate between manipulatedandgenuine photos generated by modern techniques such as face2face, deep fake, andface swap. Current methods require substantial data input, high computational power, and are time-consuming. To overcome these challenges, we propose a novel HybridLearning model that ranks features to detect fake and genuine images usinglesscomputational power and time, while improving accuracy to 99 %and MCCscoreto98.87 additionally reducing computing time by 40 seconds.