Image Information Hiding Method Based On Deep Learning

The image information hiding method embeds the secret data into the digital image and transmits it via the public channel,so that the unauthorized party cannot detect it,and can realize the transmission of secret information.However,due to the improvement of computer computing power,traditional information hiding methods cannot guarantee the secure communication of information.In recent years,the rapid development of deep learning technology and the image information hiding method designed in combination with deep learning can effectively resist the attack of the steganographic model and prevent the leakage of secret information.Aiming at the problems of image distortion in steganographic images and easy to be detected by statistical feature analysis,in this thesis,we use the adaptive quantization matrix coding and region feature proposal steganography model to embed the secret information,so that the generated secret image has a high concealment,and the secret information can be completely restored,which improves the security of the secret communication and has a high application value.Use adaptive quantization matrix coding to embed the secret information into the secret image.Firstly,the image is divided into non-overlapping regions of the same size,and the complexity of the image is defined by DCT transformation;secondly,the region blocks are converted into vector representations by column stacking,and the quantization matrix parameters are defined by the vector groups of different coefficient sizes of the image,and the design adaptive quantization matrix;finally,the secret information is mapped to the image linear quantization sequence with smaller coefficients through the adaptive quantization matrix coding algorithm to complete the embedding of the secret information.The image encoded by the adaptive quantization matrix is embedded into the carrier image through the regional feature proposal steganography model.By improving the structure of the Inception module,using convolutional stacking to construct a new network model,quickly extract the multi-dimensional feature values of the image and construct a new feature map.Combining the regional proposal network to determine the proposed area where the feature value is embedded in the carrier image,use Ro I pooling to extract and classify each proposed area,select the location in the carrier image suitable for embedding the feature value,and generate the steganographic image.The secret image is extracted through the similar structure of the network model,and finally the secret information is restored by the adaptive quantization matrix decoding.The experimental results shows that the image information hiding method based on deep learning proposed in this thesis has high embedding rate,low bit error rate,good model convergence effect,high security and concealment;image feature embedding speed and extraction speed are fast,It can suppress the occurrence of distortion;the generated steganographic image is of high quality,can resist the detection of steganalysis technology,and can effectively guarantee the safe communication of data on the public channel.