Hybrid dense net 201 with CBPN based image pixel enhancement and optimized ADBGAN for image copy-move forgery detection

The rise of image manipulation, driven by advanced editing tools, has led to malicious activities like defamation and extortion. One common type, copy-move forgery, involves duplicating and repositioning elements within an image. This emphasizes the need for reliable forgery detection systems. However, current methods often face challenges in accurately detecting tampered areas, especially in large or low-contrast images, several geometrical attacks like scaling, noise addition and rotation. Using deep learning techniques, this study offers an efficient way to detect copy-move image forgeries. The process begins by selecting an image from a copy-move forgery dataset. In pre-processing, the Frost filter and hybrid DenseNet-201 combined with the Compact Back-Projection Network are used to eliminate noise from the original images and enhance the contrast of the pixels in the denoised images. The Context Encoder decoder Network technique is used to segment the preprocessed image. The CapsNet approach extracts features from segmented images, reducing duplicated gradients and focusing on the copy-move forgery area. Adaptive Drop Block-enhanced Generative Adversarial Networks detects copy-move forgery images, while the Sooty Tern Optimization Algorithm tunes hyperparameters like learning rate and batch size to enhance performance. In addition, the effectiveness of the proposed model is evaluated using metrics like a testing accuracy of 97.39%. The experimental results demonstrate that the Adaptive Drop Block-enhanced Generative Adversarial Networks model outperforms existing ones in accurately identifying the tempered area, even with noisy images.