Construction of Perceptual Security Dataset for Video Selective Encryption Based on Double-Blind Subjective Experiment

Abstract

The development of information technology has provided opportunities for the application of Ultra High Definition (UHD) video, but the security problems such as privacy leakage cannot be ignored. Video selective encryption algorithm is an efficient and effective way to protect the privacy of users. However, the previous studies on the performance evaluation of video selective encryption scheme are insufficient. PSNR and SSIM, which are commonly used to evaluate the image quality, cannot fully reflect the encryption effect of selective encryption algorithm, which makes it difficult to evaluate and compare the security protection ability of selective encryption scheme for video content. To build an accurate security evaluation scheme of video selective encryption algorithm needs to rely on a large number of encrypted video data, but the existing datasets have the problems of small number of videos, single video resolution and selective encryption method, and the subjective experiment scheme is lack of scientific. In order to solve this problem, this paper selects a variety of resolutions and common scene video sequences, and uses different selective encryption schemes to encrypt them, so as to build a dataset for selective encryption. The method of double-blind subjective experiment provides a real and accurate subjective score for the perception security and visual quality of the video in the dataset. The subjective score shows that the dataset covers the encryption sequence of each security level, which can provide an effective reference for the follow-up research on video selective encryption and perceptual security index.