Convolutional block attention based network for copy-move image forgery detection

Computer-generated picture forgery is a growing problem due to the development of easily available technology that makes the forging procedure very simple. In response, numerous methods have evolved for detecting computer-generated forgeries. This paper introduces a new AI algorithm using the deep learning concept for advanced copy-move image counterfeit detection and localization. In this work, feature extraction, segmentation of the image, and localizing the area of forgery in an image have been performed using the Convolutional Block Attention Module (CBAM). Specifically, spatial and channel attention features are fused by the convolution block attention mechanism to fully capture context information, and enrich the representation of features. Furthermore, deep matching is used to compute feature map self-correlation, and Atrous Spatial Pyramid Pooling (ASPP) is used to fuse the scaled correlation maps to construct the coarse mask. Finally, bilinear upsampling is done to resize the predicted output to the same size as the original image. The CoMoFoD dataset is used for conducting and checking the effectiveness of the proposed work. Various performance analyses conducted on the proposed work demonstrate that CBAM has superior performance for forgery detection and localization than several state-of- the-art methods and has high strength in post-processing operations, such as noise addition, noise blur, brightness change, colour reduction, and JPEG recompression.