Detection of Image Recognition Forgery Technology under Machine Vision

Traditional image forgery detection methods have difficulty keeping up with the development of forgery technology and cannot effectively detect complex forged images. With the comprehensive use of modern computer vision and deep learning technology, this paper provides a new solution for the complex image forgery problem on the internet. By improving the Xception model, the accuracy of forged image detection can be improved, and the spread of forged images can be effectively identified and prevented. First, image forgery was carried out via forms such as the generative adversarial network (GAN), the cascaded refinement network (CRN), and implicit maximum likelihood estimation (IMLE). Second, this paper preprocessed the collected forged image dataset, extracted texture features, edge features, colour features and local discontinuities of the image and performed feature-level fusion of different types of features. Last, an improved Xception model was utilised to detect forged images. The experimental results showed that homologous data detection accuracy and single heterosource data detection accuracy were 97.8% and 86.9%, respectively, in the ProGAN dataset. The improved Xception model can effectively improve the accuracy of forged image detection and provide an effective detection method for complex forged images on the internet.