A Novel Dense-swish-cnn With Bi-lstm Framework For Image Deepfake Detection

Abstract

The Rapid Advancement Of Deep Generative Models, Particularly Generative Adversarial Networks (GANs) And Diffusion-based Architectures, Has Substantially Lowered The Barrier To Producing Photorealistic Synthetic Human Faces, Collectively Referred To As Deepfakes. Such Media Present Critical Societal Risks Encompassing Identity Fraud, Large-scale Misinformation, And Coordinated Cybercrime. Existing Detection Approaches, Predominantly Convolutional Neural Network (CNN)-based Architectures, Demonstrate Adequate Performance On Benchmark Datasets; However, They Are Limited In Their Capacity To Jointly Model Spatial Artifact Patterns And Sequential Feature Dependencies Inherent In Manipulated Imagery. This Paper Proposes A Novel Hybrid Deep Learning Framework—the Dense-Swish Convolutional Neural Network Integrated With A Bidirectional Long Short-Term Memory (Bi-LSTM) Network—designed To Overcome These Limitations. The Proposed Architecture Leverages DenseNet121 As The Backbone For Dense Multi-scale Spatial Feature Extraction, Augmented By The Swish Activation Function To Improve Gradient Propagation And Representational Capacity. Extracted Feature Maps Are Spatially Reshaped Into Sequential Vectors And Processed By A Bi-LSTM Module That Captures Bidirectional Contextual Dependencies, Thereby Enhancing Discriminative Power Against Sophisticated Forgeries. Empirical Evaluation On A Curated Real-and-fake Image Dataset Yields A Classification Accuracy Of 99.37%, Precision Of 99.44%, Recall Of 99.31%, And F1-score Of 99.37%, Representing Consistent Improvements Over CNN-only, DenseNet Transfer Learning, And Dense-Swish-CNN Baselines. Deployment Is Realized Through A Flask-based Web Application Supporting Real-time Image Upload And Classification Inference.

Keywords