High-Pass Filter Assisted Deep Feature Extraction For Image Forgery Detection

With the increasing use of image editing techniques, ensuring the authenticity of digital images has become a critical challenge. Image inpainting, often used for object removal and restoration, can also be exploited for malicious purposes, making its detection essential in forensic applications. This paper presents a deep learning-based approach for image manipulation detection, which uses a combination of high-pass filtering and convolutional neural networks to enhance feature extraction. The proposed model consists of three primary blocks: an enhancement block, an encoder, and a decoder. The enhancement block applies a high-pass filtering layer to capture fine-grained texture details and highlight inconsistencies in manipulated images. A ResNet-based encoder extracts hierarchical features, while a decoder with transposed convolution layers reconstructs the inpainted mask. The model is trained and evaluated on a dataset containing images manipulated by various inpainting techniques. The experimental results indicate high detection accuracy, with significant improvements in recall and F1 score, demonstrating the model's ability to identify manipulated regions effectively. This research advances image forensics by improving automated inpainting detection, offering a promising direction for real-world applications in digital forensics, media authentication, and security.