Professional Documents
Culture Documents
APPLICATIONS
Azeddine Beghdadi⋆ , Muhammad Ali Qureshi† , Borhen-eddine Dakkar⋆ , Hammad Hassan Gillani† ,
Zohaib Amjad Khan†† , Mounir Kaaniche⋆ , Mohib Ullah‡ , Faouzi Alaya Cheikh‡
⋆
University Sorbonne Paris Nord, France
†
The Islamia University of Bahawalpur, Pakistan
††
L2S, CentraleSupélec, University Paris Saclay, France
‡
Norwegian University of Science and Technology (NTNU), Norway
Rain: Significant intensity variations in videos due to rain cording to the importance of one distortion component over
may degrade the performance of outdoor surveillance. The the others. It is worth noticing that the order of application of
generation of videos containing rain at different density lev- the different synthetic distortions must take into account the
els is also obtained by following the same process as in the physical reality and in particular the way in which the signal
case of haze and smoke. By varying the α parameter values is acquired by the sensors in real scenarios, e.g., if we gener-
we generated different levels of rain density. ate noise on a video and then apply the blur, the effect of noise
Compression Artifacts: Various artifacts due to the com- would be affected by lowpass filtering effect due to blurring.
pression or the transmission channel can affect the quality of Figure 1 (c) presents examples of video frames with multiple
the videos. In this study we consider mainly the blocking distortions.
artefact resulting from FFmpeg coding at four quality levels.
Figure 1 (b) presents video frames with single semi-artificial
4. DATASET CHARACTERISTICS ANALYSIS
distortion.
It is important to have different scenarios but also visual con-
3.3. Multiple Distortion Generation tents rich in spatio-temporal structures at different scales of
observation and under various lighting and viewing angles.
The process of generating mixtures of simultaneous distor- The richness or diversity in videos is computed through spa-
tions is purely intuitive and is based on an additive or mul- tial and spatio-temporal descriptors. A set of criteria and mea-
tiplicative model where the different terms are weighted ac- sures to quantify and analyze the richness and representative-
(a) SI versus CF (b) GCF versus CF (c) TI versus SI
Fig. 2. Scatter plots for VSQuAD. The symbol ’⋆’ represents the keyframes of reference videos.
ness of image and video databases dedicated to perceptual frame. The maximum value over time is used to represent the
quality assessment has been proposed in [15, 16]. In what spatial information content of the scene. The SI measure is:
follows we will recall some of these descriptors and apply
them to the analysis of the database. SI = maxn {std[Sobel(Fn )]} (2)