Forgery localization in images

• Main Author: Nur Dilah Binte Zaini
• Other Authors: Deepu Rajan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/171981
• Institution: Nanyang Technological University

Due to the increase in popularity for forgeries in images, verification of authenticity in images techniques are required. Manipulation techniques are categorized into 2 categories, information-changing techniques and information-preserving techniques. In this project, a universal feature set is created to detect 10 information-preserving manipulation types in regions of image. The proposed information-preserving forgery localisation method is training a simple supervised machine learning model with the universal feature set to identify forged regions in post-JPEG compressed images. Forgeries in images over the past years have been evolved to deep fakes. To combat the creation of deep fakes, many deep fake detections have been implemented to determine the authenticity of the image. The proposed deep fake detection method uses facial parts which are eyes, nose and mouth as image patches to classify the entire image as ‘real’ or ‘fake’. Each facial part is used to train a random forest classifier respectively. Late fusion is implemented to combine the confidence scores of the predicted class for each classifier. Simple machine learning methods have been carried out to implement image forgery detection and deep fake detection in this paper. Experiments are conducted to retrieve the classification accuracy in both proposed detectors.