A HEVC Video Information Hiding Algorithm Based On Deep Learning

Steganography is designed for secret communication.As people become more and more fond of information exchange by a video,the information hiding based on video will have a huge application demand in copyright protection,military communication and other fields.The amount of data in the raw video is huge,so almost all the videos we usually see are compressed.Since its release in 2013,HEVC(High Efficiency Video Coding,HEVC)has received widespread attention for its high compression efficiency,and it has been applied in many products after recent years of development.Therefore,it is of great significance to study video information hiding algorithm based on HEVC.This thesis proposes a high-performance HEVC video steganography algorithm based on PU(Predict Unit,PU)partition modes.According to different information embedding levels,it can meet different requirements for transmitting secret information.The proposed algorithm takes the PU partition modes of 8×8 and 16×16 CUs as the carrier to embed information.The diamond coding is introduced to effectively improve the information expression ability of limited types of PU modes,and finally 1.79 bits and 2.32 bits of binary information can be respectively embedded in an 8×8 and 16×16 CU.Compared with the latest and classic HEVC video steganography algorithms,the proposed steganography algorithm greatly increases the embedding capacity while controlling the impact on visual quality and bitrate to a lower degree.In the field of video information hiding,embedding capacity,visual quality and bitrate change are the three basic indicators to evaluate the performance.During the process of information embedding,the optimal compression parameters selected by the default HEVC would be modified,which has a negative impact on the visual quality and bitrate of the compressed video.As far as we know,this thesis firstly introduces CNN(Convolutional Neural Network,CNN)into the field of information hiding to solve the basic contradiction among embedding capacity,visual quality and bitrate.Experimental results show that,compared with the default HEVC,the PSNR(Peak Signal-to-Noise Ratio,PSNR)of steganographic videos is reduced by an average of0.06 d B and the bitrate has increased by 3.47% after information embedding.The introduction of CNN has effectively improved the visual quality of steganographic videos,and even the average PSNR of the steganographic videos is 0.16 d B higher than that of the default HEVC video,and the bitrate increase has been reduced from 3.47%to 2.98%.It can be seen that the introduced CNN effectively improves the visual quality and reduces the bitrate of the steganographic video without affecting the embedding capacity,which opens up a new idea for improving the overall performance of the steganography algorithm.Finally,this thesis further evaluates the security and computational complexity of the proposed algorithm.The experimental results show that the proposed steganography algorithm has a strong anti-steganalysis ability for the two detection algorithms based on the PU partition mode,and it has lower computational complexity.