Enhancing DenseNet networks for improved forensic image analysis and tampered image detection

The increasing sophistication of image manipulation techniques challenges traditional forensic image analysis (FIA) methods. Detecting tampered images accurately and efficiently has become crucial, particularly in sectors like law enforcement and media. The objective of this research is to enhance DenseNet architectures to improve tampered image detection by increasing accuracy, reducing processing time, and improving robustness. The approach combines advanced techniques, including Gabor-bilateral filtering (G-BF) for improved feature extraction, MS-DenseNet for multiscale feature extraction (MSFE) and attention mechanisms (AMs), and GAN-DenseNet to generate realistic features. These methodologies help address limitations in detecting subtle image tampering. Enhancing DenseNet improved tampered image detection accuracy from 85% to 95% and reduced processing time from 5 to 7 s to less than 1 s. The model also demonstrated increased robustness, making it suitable for real-world applications in forensic analysis. The future work aims to integrate advanced AMs, fine-tune GANs to enhance feature generation by 10%-15%, optimize real-time detection, and target 98% accuracy for further advancements in FIA and tampered image detection.