부산대학교 도서관

Most existing methods for hiding data depend on the features present in each image or frame, so these characteristics will be lost when exposed to any attack. For this reason, in this work, we present a novel method based on an activity in the video. Our approach is to send a secret message using a de Bruijn sequence formed as voluntary blinks. The sender converts a secret text message into binary blocks using a Huffman coding tree and then builds a mapping table based on specified de Bruijn sequence dB(2,5) to create an index key which is needed at the receiver side to recover the secret message fragments, then the sender records a video of a person blinking voluntarily, long and short blinks, with the corresponding $0^{\prime}$s and $1^{\prime}$s in dB(2,5). The receiver reads stego-video frames, then for each frame, face detection and eye localization processes will be applied based on the mediapipe model, then the eye region passes into CNN to determine the state of the eye. Thus, define the type of blink if it is long or short, then the receiver translates long and short blinks into $0^{\prime}$s and $1^{\prime}$s in order to produce the sent dB(2,5). The experiment results showed that the accuracy of extracted secret bits was very high, reaching 100% if blinks were sent without any errors. As well, the experiments showed that the robustness reached 95.2%, which means the proposed method has more robustness against salt and pepper, Gaussian, speckle, cropping, rotation, and translation attacks) than other coverless data hiding techniques.