CABAC-Based ROI Encryption with Mask R-CNN for VVC Codec

Sio Kei Im, Ka-Hou Chan

Abstract: With the higher resolution coding of the VVC codec, the protection of private content must be considered in social contexts. Traditional encryption methods always encrypt the entire frame, removing all non-private content that is not unfavourable for compression coding. To overcome this problem, this work investigates the ROI approach, which allows the extraction of areas that contain some private visual data, thus allowing a trade-off between privacy and encryption performance. Moreover, the non-square coding block supported in VVC provides a more flexible block structure for ROI selection in video. This work presents a coding block ROI encryption method based on CABAC in the VVC framework. The block to be encrypted is determined by Mask R-CNN approach, which achieves interesting object determination and passes the coding block containing private content to CABAC coding. The proposed encryption algorithm is integrated into the CABAC processing, which makes use of the CABAC variable to enhance protection and ensure compliance with the final bitstream production. Experimental results report that the proposed encryption algorithm performs a secure protection of privacy in the VVC video content.