International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 09 | Sep 2024
www.irjet.net
p-ISSN: 2395-0072
Improving Digital Image Forgery Detection through Transfer Learning Techniques Panchami B R, Dr Prasad G R Department of CSE, BMS College of Engineering, Bengaluru, India Professor, Department of CSE, BMS College of Engineering, Bengaluru, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In today's digital age, images shared on social
The objectives section details the aims and targets of the research endeavor. It outlines key goals such as advancing the methods for detecting image forgery, enabling the simultaneous identification of various types of forgery, and applying deep learning techniques to boost accuracy.
media have become a primary means of communication. However, the rise of sophisticated malicious software capable of falsifying images has made it essential to identify such forgeries. Existing research in this field often focuses on detecting a single type of forgery, such as image splicing or copy-move, which may not be applicable to real-world scenarios. This paper introduces a novel method for enhancing digital image forgery detection by leveraging deep learning techniques and transfer learning. The proposed approach aims to identify two types of image forgery simultaneously by examining the compressed quality variations in forged regions compared to the rest of the image. The method involves creating a feature image from the difference between the original and compressed versions of the image. This feature image is then used to train a pre-existing deep learning model, which has had its original classifier removed and replaced with a new, fine-tuned one. The study evaluates eight different pre-trained models adapted for binary classification. The experimental findings demonstrate that the proposed method, when applied to these eight models, significantly outperforms current state-of-the-art techniques, as evidenced by various evaluation metrics, charts, and graphs. Notably, the DenseNet121 model achieved the highest detection accuracy (approximately 98%) with fewer training parameters, resulting in a quicker training process.
Additionally, this section specifies the goal of comparing the performance of different pre-trained models to gauge their effectiveness in detecting forgeries. It also emphasizes the objective of demonstrating the benefits of incorporating these advanced techniques into forgery detection strategies.
2. RELATED WORK In reference [1], the authors introduced an innovative method for detecting Copy-Move Forgeries (CMFD) by integrating deep learning techniques, specifically using a combination of Convolutional Neural Networks (CNNs) and Convolutional Long Short-Term Memory (ConvLSTM) networks. Their approach involves extracting features from images through a sequence of convolutional layers, ConvLSTM layers, and pooling operations, followed by a feature matching process to identify forgeries. The method was evaluated using four public datasets: MICC-F220, MICCF2000, MICC-F600, and SATs-130, which were merged to form new datasets aimed at improving generalization and reducing overfitting. Additionally, the performance of a ConvLSTM-only model was assessed to compare with the hybrid ConvLSTM-CNN model. The results demonstrated that the proposed method achieved exceptional accuracy, reaching up to 100% on some datasets, with processing times as brief as 1 second, thereby surpassing the performance of previous methods.
Key Words: Detect forgery, Deep learning, DenseNet121, MobileNetV2, Image Forgery, CNN, Transfer Learning etc.
1.INTRODUCTION In the current digital era, image forgery has become a significant issue, particularly with the widespread use of social media and online platforms. The manipulation of images has become increasingly prevalent, leading to concerns about misinformation and the spread of fake news. This underscores the critical need for effective forgery detection methods to address these challenges. Advanced editing tools have made it easier to alter images, complicating the task of identifying various forms of forgery. As a result, developing robust detection techniques is essential for ensuring the integrity of visual information and combating the negative effects of digital deception.
© 2024, IRJET
|
Impact Factor value: 8.315
In reference [2], the authors present a technique for identifying both copy-move and splicing image forgeries using Convolutional Neural Networks (CNNs) and three distinct models: Error Level Analysis (ELA), VGG16, and VGG19. Their method includes a pre-processing phase where images are compressed to a particular quality level before being used to train the models. Once trained, these models are employed to classify images as either genuine or manipulated. The experimental findings indicate that the proposed approach yields accuracy rates of 70.6% with Error Level Analysis (ELA), 71.6% with VGG16, and 72.9% with VGG19, when tested on images from the CASIA2.0 and NC2016 datasets.
|
ISO 9001:2008 Certified Journal
|
Page 769