Quality assessment of
encoded sequences subject to transmission losses
The Absolute Category Rating (ACR)
subjective test methodology has been used.
Each participant is presented
a set of video sequences. After the presentation of each video
sequence, he is asked to score its quality - see
for additional details.
53 individuals from both
genders participated in this test. Their ages were in the range 17 to 50
(most of them were students).
11 participants declared to be
familiar with video encoding processes but none of them was
working in the video quality assessment field.
Participants were screened for normal
visual acuity and color blindness. Their assessments were
analyzed in order to detect inconsistencies.
The above procedures detected 3
blind color persons and 2 participants with inconsistent
results. Therefore, a total of 48 participants were considered
for computing MOS values.
The tests were performed at two different rooms which were
configured in order to achieve identical illumination
Displays of the same model
(ASUS ML238 LED-LCD) were
calibrated to the same settings and
used in all test
were as listed below:
- Height of the picture shown in the
screen: 14 cm
- Viewing distance: approx. 65 cm
- Background room illumination:
- Peak luminance of the LCD screen:
- Ratio of luminance of inactive
screen to peak luminance: 0.023
- Ratio of luminance of background
behind the display to peak of luminance: 0.05
original sequences were encoded using the
H.264/AVC JM 14.2 reference software.
The resolution of all sequences is 720x480 and their
duration is 10 seconds (length of 300 frames for 30 Hz
sequences; 250 frames for 25 Hz sequences)
The main encoder parameters were as follows:
- Main profile, level 2.0
- GOP structure: 15 frames length, with two B frames
inserted between I/P frames
- One IDR frame every 1 second of video
- Maximum of 5 reference frames used for motion estimation
- Low complexity RD optimization mode
- Original JM rate control algorithm
- RTP output format - one video slice per packet
- 45 macroblocks per slice (each slice is a complete row
The encoded sequences have been corrupted by simulating the
transmission of the streams over an error-prone network.
The simulation was carried out using
the software associated to
Rec. ITU-T G.1050 - "Network model for evaluating multimedia
transmission over Internet Protocol".
The generated loss patterns correspond to packet loss rates
ranging from 0.17% up to 7.90%.
All bitstreams are available in the following links. In
order to avoid large file sizes, they have been grouped in the
All video streams can be decoded using the batch file and the
H.264 decoder* provided here (place the contents of the zip
file in the same directory as the streams).
Filenames format used in the streams:
- Encoded sequences: <video sequence>_<spatio-temporal_resolution>_<bitrate>
- Corrupted sequences: <video sequence>_<spatio-temporal_resolution>_<bitrate>_corrupted_plr_<packet
* Note: the decoder provided in the zip file is a modified
JM H.264 decoder that allows the specification of the NALU
type (Annex B / RTP format) in the command line. It is the
same used in the
Polimi-EPFL video quality assessment database.
Resulting MOS values and their standard deviations can be downloaded in excel or text file
- MOS (text file)
- each line contains the data for a test condition,
organized as follows: sequence name, MOS value and MOS
standard deviation, separated by tabulations.