Research On Source Camera Identification Algorithm Based On Color Correction Matrix Feature

This thesis focuses on source camera model identification,and a new algorithm based on image color correction feature is proposed on the basis of previous studies.A new convolutional neural network structure is designed for the extraction of image color correction feature and the identification of source camera model,and the algorithm optimization is designed for multi-scene problems.The main researching content of this thesis includes the following aspects:Firstly,this thesis proposes a new feature,image color correction(CCM)feature,that can be used for source camera model identification.By extracting the color correction matrices of different types of device and performing correlation analysis,this thesis obtains the conclusion that the image CCM feature is unique on different types of devices,and then verifies its effectiveness in the field of source camera model identification.Secondly,this thesis designs a source camera model identification network CCMNet based on image CCM features.Through the improvement of the fully connected layer mapping structure and loss function,the network can extract the CCM features of the image itself,and compare them with the CCM real value library of the device in order to complete the task of camera source model identification.Finally,this thesis designs a multi-scene feature fusion identification network CCM-Ms Net.By using the improved multi-scene attention mechanism module,the network can adaptively adjust the receptive field and pay more attention to the key feature areas,so as to realize feature fusion at different scales and complete the camera source identification task close to the real scene.In this thesis,the experimental verification is carried out on a large-scale data set.In the singlescene condition,the camera source model recognition accuracy of CCMNet can reach up to 97.23%,while the CCM-Ms Net achieves accuracy of 97.15% under multi-scene condition.The performances of these algorithms are both superior to other current mainstream algorithms.The experimental results show that the image color correction feature has high research value and application potential in the field of source camera source model identification.