| Image hiding is a technique to store and spread secret information in images.It is widely used in information transmission,copyright protection and other fields.At the same time,the secret image is imperceptible and can be recovered safely at the receiving end.The current hiding algorithms have low hiding capacity,weak ability to resist steganalysis and poor robustness,and it is difficult to achieve a good balance among various performance indicators.In view of the shortcomings of existing algorithms,this paper focuses on improving the hiding capacity,security and robustness of image hiding algorithms,and has achieved the following research results:Firstly,in order to improve the hiding capacity,an image hiding algorithm based on generating feedback residual network(GFR-Net)was constructed.The algorithm hided one or more color secret images in a color carrier image to obtain the container image.A recovery network GFR-Net was designed to reconstruct the secret image from the container image.Experiments show that the proposed image hiding model based on GFR-Net had good performance in terms of payload and security.Secondly,in order to overcome the security problems caused by high capacity,an image hiding algorithm based on wavelet domain and residual convolution neural network(Res CNN)was proposed.First,the low-frequency subband of the secret image after wavelet decomposition was discarded,and three high-frequency subband were retained and spliced together as the characteristics of the secret image.Subsequently,these features were effectively embedded into the carrier image to obtain the container image.Res CNN can effectively extract hidden features from containers and reconstruct secret images.Experimental results show that the proposed image hiding algorithm is effective and achieves the best results in terms of hiding capacity and security.Thirdly,in order to improve security,the secret image was encrypted and then embedded into the carrier image.The encrypted secret image reduces the correlation of pixels and reduces the influence of the carrier.During training,different weights were used for the loss function of hiding and recovery network,which made the loss function of hiding network more weighted,sacrificing the performance of partial secret image recovery,and bringing higher security.Torch-net was specially designed to recover secret images from containers.Experiments show that this algorithm can effectively resist the detection of steganalysis algorithms.Fourth,in practical applications,container images will be subjected to destructive attacks such as cutting,compression and noise pollution during transmission,affecting the recovery of secret images.Therefore,an image hiding algorithm with high robustness based on attention mechanism,residual and convolution neural network(ARes-CNN)was proposed.First,the secret image was decomposed by Haar wavelet transform.After an eight-level decomposition,the low-frequency sub-band degenerated into one pixel.The reduced low-frequency sub-band information can improve the security of the hiding algorithm to resist various steganalysis attacks.These 25 sub-bands were spliced into a whole,and then embedded into the carrier image through the hidden network.The experimental results show that even if the container image is attacked destructively,it can recover the secret image with high quality,and the algorithm is highly robust.The capacity was chosen as the starting point,and GFR-Net was constructed to realize the image hiding of large capacity,while the security and robustness were ignored.In order to improve security,the secret image was embedded into the carrier image after wavelet transform,effectively improving the quality of the container image.When the embedding capacity reaches 3200%,a higher quality container image can still be obtained.However,embedding too much information can cause direct interference with each other,and the recovery network can not effectively recover the secret image.The hiding algorithm in the wavelet domain has improved security,but the improvement was limited.In order to further improve the security,the secret image was encrypted to hide,which has been effectively improved in security.In the ability against steganalysis,it was superior to other advanced hiding algorithms,but its robustness remained a problem.Based on the above three algorithms,a new algorithm was proposed in the wavelet domain,which achieved a relative balance in terms of concealment,capacity,security and robustness.Research results can be directly applied to secret communications,medical diagnosis,judicial forensics and other fields,and will be of important significance in application. |
