Impact of Deep Learning Techniques on Deep Fake Image Identification for Digital Investigation

In recent years, deep fakes have become increasingly prevalent and sophisticated, raising concerns about misinformation. This has led to a surge in research on deep fake creation, detection techniques, and related datasets. The best approaches for detecting deep fakes often involve deep learning techniques, primarily using convolutional neural networks (CNNs). This article proposes a deep learning-based method using CNNs to identify deep fake videos by analyzing frame-level inconsistencies that generative models struggle to replicate. Our approach combines multiple convolutional layers to capture spatial hierarchies and detect subtle visual cues that differentiate real from fake content. The proposed model demonstrates strong performance, achieving high detection accuracy and robustness across several deep fake generation methods. Extensive testing on public deep fake datasets shows that CNN-based models are effective tools for automated deep fake detection, offering a promising countermeasure to the growing misinformation problem. In Deep Fake and Real image (Dataset1) and Bigger Dataset for Image Deepfake Detection (Dataset2), these two dataset we used to detect the real and fake images and the highest accuracy of dataset-1 using CNN model is 96.28% among all the models and in the dataset-2 the highest accuracy was 68.02% by using the Xception model. Following that, we built and trained our model with the maximum accuracy at an acceptance rate of 96.2% using CNN,NN, Xception, DenseNet-121, and EfficientNetB0.