Research On Selective Encryption Of ROI Based On H.265/HEVC Video

Videos widely circulated on public networks inevitably contain privacy information,such as the portraits of individuals in video conferences,pedestrians in surveillance videos,or interview subjects in public media.The areas containing these objects are referred to as privacy regions.Consequently,numerous encryption algorithms for videos have been proposed.Some video encryption algorithms directly encrypt the raw data of the video sequence or the encoded video stream data.However,the former disrupts the video’s correlation,leading to a significant decrease in coding efficiency,while the latter compromises the format compatibility of the video stream.In order to ensure the format compatibility of the video stream,selective encryption algorithms for videos have been proposed.However,most of these algorithms encrypt the selected syntax elements of the entire video frame,rarely distinguishing between privacy and non-privacy regions.This results in higher computational complexity and provides limited detailed schemes for privacy region localization and decoding/decryption.Therefore,this to some extent limits the practical application of video encryption algorithms.To address the issues in existing video encryption algorithms,this thesis focuses on using the human face as the privacy region,namely the Region of Interest(ROI),and combines it with the self-synchronizing chaotic stream cipher algorithm to study selective encryption of High Efficiency Video Coding(HEVC)video ROI.The main contributions of this research are as follows:(1)In order to obtain the pseudo-random sequence required for encryption,this thesis designs a ciphertext feedback-type self-synchronous chaotic flow cryptosystem.The pseudorandom sequence generated by the system is closely related to the plaintext and can effectively resist the selective ciphertext attack and the conquest attack separately.Furthermore,the selfsynchronous chaotic flow cryptosystem is applied to image encryption,which verifies the validity and security of the pseudo-random sequence generated by the system.(2)To address the ROI positioning problem,this thesis utilizes the Tile mechanism in H.265/HEVC,dividing video frames into multiple Tiles and transforming ROI localization into Tile localization.After extracting the images of key frames(I frames)from the video,the Open CV facial recognition module is employed to detect the rectangular bounding boxes corresponding to faces in the current image as ROIs.Utilizing the motion vector(MV)information in the coding units,the ROI of the I frame is obtained,and motion tracking is performed on the ROIs of the forward predicted frames(P frames)by calculating the polar median of MV.To enable the decoding end to obtain the ROI’s position information,this thesis utilizes the Supplemental Enhancement Information(SEI)mechanism in HEVC to transmit the encrypted Tiles’ index data.This data is attached to the Network Abstraction Layer(NAL)of the HEVC video stream and,after network transmission,the decoding end can obtain the Tile data by parsing the SEI,thus acquiring the ROI’s position information.Experimental results demonstrate that the proposed method can achieve ROI detection and tracking,reducing ROI positioning time and improving coding efficiency compared to traditional per-frame detection approaches.(3)In the selective encryption scheme for the ROI contained in the list of encrypted Tiles,this thesis uses the motion vector difference(MVD)symbol bits,MVD values,and quantized transformation coefficient(QTC)symbol bits employed in bypass encoding mode during entropy coding as encryption syntax elements.Subsequently,the proposed encryption and decryption algorithms are embedded into the official H.265/HEVC codec core,HM16.9.Multiple video sequences from international standard video libraries are used for experimental validation,and the performance of the encoder after embedding the encryption algorithm is analyzed.The experimental results show that the proposed encryption scheme has good encryption effectiveness for the ROI.Performance analysis reveals that after embedding the proposed algorithm,the average increase rate of video stream data volume is only 1.78%,and the average increase rate of encoding time is only 1.67%.Therefore,the proposed method can ensure both security performance and coding efficiency,demonstrating certain practical application value.