Multimedia Signal Processing and Security Lab

Paper

Heinz Hofbauer, Florent Autrusseau, Andreas Uhl
‟Quality and Recognizability Estimation Video Encryption Database”
Published in Information Sciences, 2024.

USEEVid DB Download

Terms of Use

The database is provided as is. If you use the database cite the following paper with which the database was published:

‟Quality and Recognizability Estimation Video Encryption Database”, by Heinz Hofbauer, Florent Autrusseau, Andreas Uhl. Published in Information Sciences, 2024.


        @article{Hofbauer24a,
            author     = {Heinz Hofbauer and Florent Autrusseau and Andreas Uhl},
            title      = {Quality and Recognizability Estimation Video Encryption Database},
            journal    = {Information Sciences},
            volume     = {678},
            pages      = {17},
            year       = {2024},
            issn       = {0020-0255},
            doi        = {10.1016/j.ins.2024.120919},
            url        = {https://www.sciencedirect.com/science/article/pii/S0020025524008338},
        }

Contact:

Heinz Hofbauer (Maintainer)

Description:

This package contains the database of videos with quality and recognizability scores by human observers.The individual observer scores as well as the outlier pruned mean observer scores (MOS) are contained within.

The database is described in detail in the paper given above.

Links to relevant papers:

Directories

encrypted: Contains the encrypted video sequences.
references: Contains the reference videos sequences.
results: Contains the result of the quality/recognition acquisition.

Results

The results are given as a csv file with sequence names, MOS and individual scores. The following files are given:

qual_all.csv: All quality results as obtained, including outliers.
qual_prune.csv: All quality results with outliers pruned.
recog_all.csv: All quality results as obtained. (No outliers were found in the recognition experiments)

Reference Videos

The videos are taken from Xiph.org Video Test Media

They are down-sampled to a size of from 1080p (960x540) so two can be displayed side by side on a standard 1080p display without issue. The sequences are cut to roughly 5 seconds or 150 (with 29.97 fps) in length.

Encryption Types

H.265 Encryption based on HM

Encryption of sign coefficients, nonzero dct positions and motion vector difference signs.
Lowdelay configuration is used. Rounds of encoding and decoding are used to reduce the quality further.

Filenames: {refid}_265_{METHOD}_{PARMETERS}.yuv

refid

Id of the reference image

METHOD

Contains one or more of the following tags, all included were used.

12ac: encrypts the encryptes the first and second AC coefficient.
mvd: encrypts the motion vector difference signs.
co: encrypts all coefficient signs
nz: encrypts position of non-zero coefficients

PARAMETERS

This can either be qxx in which case q gives the quantization parameter, i.e. encryption is done a the given quality level with one round.
Or it can be in the form of yyy which refers to the rounds of encryption. In this case QP was fixed to 40.

JPEG2000 encryption based on JJ2000

Encryption of data only based on an offset and window size, encryption windows size of the bitstream starting at the offset. Depending on the encoding of the bitstream data is ordered in a different way, supported encoding types for JJ2000 are layer and resolution progressive. In our case the final encryption settings all make use of the resolution progressive ordering of the bitstream. Encryption is applied on a frame by frame basis.

Filenames: {refid}_mj2_{PROGRESSION}_o{OFFSET}_w{WINDOW}.yuv

refid: Id of the reference image
PROGESSION: Contains the progression type of the encoding, could be l for layer or r for resolution.
All our encryption settings used r in the end.
OFFSET: Offset where the encryption starts, everything up to this point is in plaintext. Generally speaking, the higher the offset the better recognizable the content of the video will be, although, this also depends on the window size.
Given in percent of the overall bitstream
WINDOW: This is the amount of the bitstream encrypted, starting at the offset, given as a percentage of the overall bitstream size.

BEWARE: The database is large (~ 9.5GB to download)
txz: This is a txz file, that is tar with xz compression. Due to the database size this compression was chosen over the more common zip (which would have resulted in a 12GB download).

Download USEEVid DB:

The data/code is available upon request:

University of Salzburg Encryption Evaluation - Video Database (USEEVid DB)