A hybrid model for image forgery detection using deep learning with block and keypoint methods

Digital images serve as crucial evidence in fields like forensics and medicine, yet their reliability is increasingly threatened by sophisticated forgery techniques such as copy-move. While traditional block-based, keypoint-based, or deep learning approaches exist, each faces challenges regarding geometric transformations, dataset dependency, or computational complexity. This paper presents a novel hybrid model, HDBK, integrating deep learning, block-based, and keypoint-based methods for forgery detection at both image and pixel levels. The proposed framework comprises three fundamental stages. First, a triple-architecture ensemble of deep learning networks identifies forgery images and generates localized heatmaps to highlight suspected regions. Next, an enhanced block-based method utilizing a Genetic Algorithm (GA) is employed. This stage uses the maximum number of matched keypoints between candidate blocks as a fitness function to identify regions similar to the suspected forgery blocks. Finally, keypoint descriptors (SIFT, SURF, and FAST) are applied to match features between the identified regions, achieving precise pixel-level localization. This hybrid approach effectively reduces the search space for optimization, enhancing accuracy while minimizing false positive rates. The HDBK model was rigorously evaluated on the CoMoFoD dataset, which includes forgeries under various geometric transformations and post-processing operations like blurring and JPEG compression. Experimental results demonstrate that the model outperforms state-of-the-art techniques, particularly in detecting challenging scenarios such as small-scale and smooth forgery regions. The synergy between deep feature maps and meta-heuristic optimization ensures a robust balance between computational efficiency and forensic integrity in real-world passive forensic applications.