Smart Detection of Deepfake Using Various Deep Learning Models

Simple item page
dc.contributor.advisorMohammad Monirujjaman Khan
dc.contributor.authorMariam Binte Bashir
dc.contributor.authorIftekher Mahbub Rafi
dc.contributor.id2021874042
dc.contributor.id2021463642
dc.coverage.departmentElectrical and Computer Engineering
dc.date.accessioned2025-04-22
dc.date.accessioned2025-04-22T04:18:41Z
dc.date.available2025-04-22T04:18:41Z
dc.date.issued2023
dc.description.abstractDeepfake is a digitally manipulated image or video that is generated using an algorithm to replace the original person with someone else which looks authentic. The need to create efficient techniques for identifying and categorizing counterfeit photos on digital platforms has become crucial due to their widespread presence. This project does the prediction of counterfeit photos by employing of deep learning (DL) and transfer learning methodologies, specifically leveraging convolutional neural networks (CNNs) models. Our work focuses to enhance the accuracy and reliably detecting deep fake images. The DL models have been trained using the dataset acquired from Kaggle, titled "140k Real and Fake Images". Transfer learning techniques are employed to enhance prediction performance by leveraging knowledge from large datasets and applying it to pre-trained models. Evaluation criteria such as accuracy, precision, recall, and F1 score are employed to assess the ability of a model to effectively distinguish between genuine and manipulated images. The models exhibit strong discriminatory capabilities and provide reliable picture classifications. The study investigates the efficacy of various deep-learning models in the task of image classification. The Vgg16 achieved a peak accuracy of 99%, showcasing the possibilities of deep learning algorithms. MobileNetV3s and DenseNet121 demonstrate their efficacy in faulty image categorization by achieving accuracies of 98%, 94%, and 95% correspondingly. Unlike other models, Xception achieves remarkable accuracy rates of 96% through training on a wide range of datasets containing both genuine and manipulated images. This enables it to effectively differentiate between real image and deepfakes, which we deployed to build a website. Thus, our experiment demonstrates the versatility of deep learning models in handling various image formats.
dc.description.degreeUndergraduate
dc.identifier.cd600000497
dc.identifier.print-thesisTo be assigned
dc.identifier.urihttps://repository.northsouth.edu/handle/123456789/1138
dc.language.isoen
dc.publisherNorth South University
dc.rights© NSU Library
dc.titleSmart Detection of Deepfake Using Various Deep Learning Models
dc.typeThesis
oaire.citation.endPage55
oaire.citation.startPage1
Files
Original bundle
Now showing 1 - 2 of 2
Thumbnail Image
Name:
600000497.Abstract.pdf
Size:
133.59 KB
Format:
Adobe Portable Document Format
Description:
Download
Thumbnail Image
Name:
600000497.pdf
Size:
1004.25 KB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
1.93 KB
Format:
Item-specific license agreed to upon submission
Description:
Download
Collections
Theses - Undergraduate