Computer-Generated Image Detection Based on Multi-Scale Feature Fusion Attention Module

As processing technology advances rapidly and image editors become widely used, computer-generated (CG) images are becoming increasingly lifelike, making it harder and harder to distinguish them from photographic (PG) images by eye. At the same time, the potential malicious uses of highly realistic computer-generated images have made the detection of the authenticity of digital images an important research field. In this study, we propose a method for detecting computer-generated images that combines multi-scale feature learning and the Convolutional Block Attention Module (CBAM). This method uses a feature transfer module and a feature fusion module, considering both the superficial content features of the image and deeper semantic features, thereby enhancing the recognition ability of computer-generated images. We have validated the efficacy of this method through lots of comparative experiments and comparisons with other models. Experimental results show, our model has good performance in CG image detection.