Research And Implementation Of Image Tampering Detection Technology Based On Deep Learning

Image tampering is an attack method that maliciously destroys the content of image information spread in the network.Through the comprehensive application of many image processing technologies,the information that the original image wants to express is tampered as the false information that is intentionally manufactured.The tampered image is used for visual deception or false manufacturing fact.In the context of the increasingly convenient network environment,images are widely used as the carrier of information to spread rapidly in the network.In some cases,images even play an important role in the judicial field and daily life as the proof of the truth.The rapid development of image tampering technology has brought great threat to the network information security.The traditional image tamper detection technology has been greatly affected by the image tampering behavior of diversified development and even in-depth learning technology.According to a certain feature detection method,there is a problem of poor adaptability to the new image tampering technology,and there are some problems of limitations and flexibility for image detection tasks with different tampering characteristics.This paper focuses on the difficulties in image tampering detection.1.In this paper,the principle of DeepFakes is deeply studied,and DFDM detection network is designed.Aiming at the new image tampering technology combined with deep learning,the detection network infrastructure is constructed by referring to the lightweight steganalysis model;the custom convolution kernel is introduced to guide the network to learn the texture difference of the image tampered area;the Max Pooling layer is used to reduce the network parameters while retaining the image texture characteristics to the maximum extent;the multi-layer perceptron(MLP)is used to classify the image samples.DFDM solves the problem that deep learning classification network can't detect the image details generated by GAN network.The simulation results on the "faceforensics++"dataset show that DFDM has better detection ability than other detection methods.2.This paper proposes an image content tampering detection network SFJNet.Combined with the key components of steganalysis mature model,the learning ability of the detection model to the deep information of the image is improved;the dual network joint detection model is designed to improve the limitations of single network detection in the spatial domain or transform domain.In view of the fact that image tampering operation may have different tampering characteristics in image spatial domain and transform domain,the dual network optimizes the spatial domain and transform domain respectively,and the transform domain network performs discrete DCT transform on the data.By designing the joint module and using the Maximum Likelihood Estimation(MLE)to make probability joint for the two networks,the spatial domain decision and the transform domain decision complement each other.The simulation results show that the Precision,Recall and Accuracy of SFJNet in CASIA dataset can exceed those of the same type of methods.3.According to the algorithm designed in this paper,the remote image tamper detection system based on Java Web is implemented.The system solves the adaptive problem and distributed request problem in the practical problem of image tamper detection,solves the limitations of traditional detection algorithm for tamper samples with different tamper characteristics by loading detection algorithm model on remote cloud server,and solves the congestion problem of multi request processing by distributed computing on remote server.