Research On Facial Image Modification Detection Based On Deep Learning

With the advancement and development of technology,digital images have become more and more popular,and various powerful image editing software has emerged,such as Adobe Photoshop,PortraitPro Studio Max,Meitu Xiuxiu,GIMP and Snapseed.The software brings people a lot of convenience,and makes it easy for photo holders to edit and modify facial digital images.The edited and modified digital images can be a feast for the eyes,and the perfect figure and the innocent skin in the images are also visually enjoyable.However,if digital images are tempered by unscrupulous people,this process will involve deception and forgery,and will also have a serious negative impact on personal privacy,social stability,national politics and economics development.Therefore,with the frequent occurrence of malicious tampering on face images,the research on face image modification detection technology seems particularly important.Based on this problem and combined with the advantages and the core position of the deep learning model,this paper proposes a method based on deep learning for facial image modification detection.The main research results are as follows:(1)Through the study of the theoretical knowledge of the Restricted Boltzmann Machine(RBM)model structure,training algorithm and parameter setting in the deep learning method,and then combined with the problem of facial image retouching and tampering The specific algorithm is implemented,and the detection of facial image modification is realized.The restricted Boltzmann function can unsupervised the feature extraction and automatically learn the image features.It can be used directly as the feature extractor of the face image,and then combined with the K nearest neighbor(KNN)classifier to classify the obtained facial image features.The experimental results are demonstrated in the three face image libraries used in this paper.The results show that under different learning rates,the method of RBM combined with KNN classifier achieves a high detection rate in face image modification detection.(2)By analyzing the related models and algorithms of deep learning,the application of Convolution Neural Networks(CNN)in image classification and recognition in deep learning is studied,and the problems of RBM training parameters and training time are too long.Based on the characteristics of local receptive field and weight sharing in convolutional neural networks,two new convolutional neural network models,Wang-Net1 and Wang-Net2,were proposed to detect tampering of facial retouching images.A lot of experimental tests have been carried out on the three face image libraries used in this paper.The exp erimental results show that the convolutional neural network model proposed in this paper not only solves the problems of excessive RBM training parameters and long time,but also the detection performance is improved to a certain extent,with strong robus tness and high recognition rate.