A Robust Image Forgery Detection and Localization Approach based on Context-Aware Attention Pooling and Convolutional Block Attention Module for Improved Detection Performance

Image manipulation technology has emerged and developed quickly, posing a threat to many facets of our society. Consequently, the identification of picture alteration has become more crucial. Though considerable progress has been made, past approaches to forgery detection did not account for the differences in the size of the tampered areas in each fake image. In this research, we argue that the primary cause of the low precision is the network’s incapacity to handle tampering regions of different sizes. We suggest Context-Aware Attentional pooling-based U-Net structures because of their simplicity in implementation, ease of integration , emphasis on feature relevance, scalability, noise reduction, and computing efficiency. It extends the capabilities of the U-Net by incorporating residual propagation and feedback, an attention gate, and Context-aware Attentional pooling (CAP) with Convolutional Block Attention Module (CBAM). The concept of channel mixing is larger in CBAM, which may indicate a more integrated method of managing spatial information and channel dependencies. In order to maximise 1 feature extraction, multiscale context understanding, and ultimately more accurate and dependable forensic analysis, spatial attention, channel attention, and Context-Aware Attentional Pooling (CAP) are integrated into image forensics. This model’s inclusion of Context-Aware Attentional Pooling (CAP) and Channel Attention (CA) improves its robustness against noise and compression, and improves detection accuracy and localisation with both global and local context, making it perform better than other state-of-the-art models. This combination is a potent method in the field of image forensics since it is very good at identifying subtle and intricate image manipulation.