Content uploaded by Yashar Deldjoo
Author content
All content in this area was uploaded by Yashar Deldjoo on May 21, 2017
Content may be subject to copyright.
MPEG7VisualFeaturesofMovieTrailers
SUMMARY
==============================================================================
This dataset provides a set of 774 low-level VISUAL features extracted from 3964 movie trailers.
The movie IDs are in agreement with the movie IDs provided by "MovieLens (ML) dataset"
(ML-20M or Full Version as in May 2017). All the movie titles, ratings and associated movie
genres and tags can be collected from the MovieLens website. We used the low-level MPEG-7
Low Level Feature Extraction by Bilkent university namely BilVideo-7 for the extraction of
1
MPEG-7 visual features from movie trailers.
INFORMATION ABOUT THE DATASET
==============================================================================
This dataset provides a set of 774 MPEG-7 low-level VISUAL features extracted from 3964
movie trailers. The data are contained in 4 files:
●MPEG7_feature_max_aggr. csv
●MPEG7_feature_mean_aggr csv
●MPEG7_feature_median_aggr. Csv
●MPEG7_feature_min_aggr. Csv
More details about the contents and use of these files are as follows. The description of each
column and each low-level visual feature is provided in the table below:
Table 1: Description of columns
Feature Type
Feature
Column title
length
Description
Numeric IDs
movieId
movieId
1
IDs of the trailers
MPEG7 Color
Descriptors
SCD
SC1, SC2, SC3, …, SC256
256
Scalable Color
Descriptor
CLD
CL1, CL2, CL3, … , CL120
120
Color Layout
Descriptor
CSD
CS1, CS2, CS3, …, CS256
256
Color Structure
Descriptor
MPEG7 Texture
Descriptors
EHD
EH1, EH2, EH3, …, EH80
80
Edge Histogram
Descriptor
HTD
HT1, HT2, HT3, …, HT62
62
Homogeneous
Texture Descriptor
1 http://www.cs.bilkent.edu.tr/~bilmdg/bilvideo-7/Software.html
DETAILS OF THE FEATURES
==============================================================================
Color Descriptors:
●Scalable Color Descriptor (SCD) is the color histogram in the HSV color space, with 256
coefficients (histogram bins).
●Color Layout Descriptor (CLD) is a very compact and resolution-invariant representation
of color obtained by applying the DCT transformation on a 2-D array of representative
colors in Y or Cb or Cr color space. CLD is described by a feature vector of length 120.
●Color Structure Descriptor (CSD) creates a modified version of the SCD histogram to take
into account the physical position of each color inside the images, and thus it can
capture both color content and information about the structure of this content. CSD is
described by a feature vector of length 256.
Texture Descriptors:
●Edge Histogram Descriptor (EHD)
describes local edge distribution in a frame. The frame
is divided into 16 non-overlapping blocks. Edges within each block are classified into one
of five edge categories: vertical, horizontal, left diagonal, right diagonal and
non-directional edges. The final local edge descriptor is composed of a histogram with 5
x 16 = 80 histogram bins.
●Homogeneous Texture Descriptor (HTD)
describes homogeneous texture regions within
a frame, by using a vector of 62 energy values.
DESCRIPTION OF FILES:
==============================================================================
MPEG7 Features extracted from the keyframes of a video file is aggregated into one single
feature vector. In order to choose the best aggregation strategy, we have experimented with
four different options, stored in the following dataset files.
●MPEG7_feature_max_aggr.csv
This file provides the MPEG7 low-level visual features
presented in Table 1, aggregated using max
strategy, i.e., each element of the feature
vector is the maximum of the corresponding elements of the feature vectors from key
frames.
●MPEG7_feature_mean_aggr.csv
This file provides the MPEG7 low-level visual features
presented in Table 1, aggregated using mean
strategy, i.e., each element of the feature
vector is the average of the corresponding elements of the feature vectors from key
frames.
●MPEG7_feature_median_aggr.csv
This file provides the MPEG7 low-level visual
features presented in Table 1, aggregated using median
strategy, i.e., each element of
the feature vector is the median of the corresponding elements of the feature vectors
from key frames.
●MPEG7_feature_min_aggr.csv
This file provides the MPEG7 low-level visual features
presented in Table 1, aggregated using min
strategy, i.e., each element of the feature
vector is the minimum of the corresponding elements of the feature vectors from key
frames.
CITATION
==============================================================================
To acknowledge the use of the dataset in publications, please cite the following paper:
Deldjoo, Yashar; Quadrana, Massimo; Elahi, Mehdi; Cremonesi, Paolo, Using Mise-En-Scene
Visual Features based on MPEG-7 and Deep Learning for Movie Recommendation, arXiv preprint
arXiv:1704.06109, 2017
ACKNOWLEDGEMENT
==============================================================================
This work has been supported by the AMAZON AWS Cloud Credits for Research program.
DOWNLOAD LINKS
==============================================================================
The dataset can be downloaded at:
●https://www.researchgate.net/publication/317038064_Mise-en-Scene_Dataset_MPEG-
7_Visual_Features_of_Movie_Trailers_dataset
There is another related dataset, that includes other types of visual features, that can be
downloaded through the following link:
●https://www.researchgate.net/publication/305682388_Mise-en-Scene_Dataset_Stylisti
c_Visual_Features_of_Movie_Trailers_description
FURTHER INFORMATION ABOUT POLIMI@RECSYS
==============================================================================
POLIMI@RECSYS is a research group at the Department of Electronics, Information and
Bioengineering (DEIB) at Politecnico di Milano, in Milan, Italy. The research focus of
POLIMI@RECSYS is mainly on Recommender Systems, in particular Multimedia Recommender
Systems. Please visit our homepage for further information: http://recsys.deib.polimi.it
Yashar Deldjoo:http://www.ydeldjoo.me
Mehdi Elahi:http://www.linkedin.com/in/mehdielahi
Paolo Cremonesi:http://www.deib.polimi.it/eng/people/details/159156