Video Classification Using Normalized Information Distance
Khalid Kaabneh
Graduate College of Computer Studies
Amman Arab University for Graduate Studies
Amman-Jordan

Abstract— There has been a vast collection of multimedia
resources on the net. This has opened an opening for researchers
to explore and advance the science in the field of research in
storing, handling, and retrieving digital videos. Video
classification and segmentation are fundamental steps for
efficient accessing; retrieving, browsing and compressing large
amount of video data. The basic operation video analysis is to
design a system that can accurately and automatically
segments video material into shots and scenes. This paper
presents a detailed video segmentation technique based on
pervious researches which lacks performance and since some
of the videos is stored in a compressed form using the
Normalized Information Distance (NID) which approximates
the value of a theoretical distance between objects using the
Kolmogrov Complexity Theory. This technique produced a
better results in reference to performance, high recall and
precision results .
Keywords— Shot-Boundary Detection, Kolmogorov
complexity, Normalized Information Distance.
Asmaa Abdullah
Zeineb Al-Halalemah
Amman-Jordan
binary strings. The Kolmogorov complexity, K(x) of x is
defined as the length of the shortest program that can produce
x.
The conditional Kolmogorov complexity of x and y, K(x | y), is
the length of the shortest program that produce x when y is
given as input.[8] The Normalized Information Distance (NID)
is an approximation of the Kolmogorov complexity using
binary data and standard compression algorithms. The distance
calculation between two images using NID is:
d(x, y) 
max{(| c(xy) | - | c(y) |), ((| c(yx) | - | c(x) |)}
max{| c(x) |, | c(y) |}
…(1)
where |c(x)| and |c(y)| is the length of the compressed data for
x and y respectively. |c(xy)| is the length of the compressed
data for the concatenation of x and y, and similarly for |c(yx)|.
Lower values for d(x,y) indicate a smaller “distance” between
the two images, indicating some similarity has been detected.
The question is whether a small distance in from this
calculation will translate into relevance between the two
images.
I. INTRODUCTION
I
ndexing and retrieval of digital video is an active research
area in computer science . The increasing availability and
use of on-line video has led to a demand for efficient and
accurate automated video analysis techniques. In videos, a shot
is an unbroken sequence of frames from one camera;
meanwhile a scene is defined as a collection of one or more
adjoining shots that focus on an object or objects of interest.
The segmentation of video into scene is far more desirable
than simple shot boundary detection [2]. This is because
people generally visualize video as a sequence of scenes not of
shots, so shots are really a phenomenon peculiar to only video.
A number of researches have been devised to implement a
segmentation systems based on in reference to image regions
and features such as color, size, shape, and others[ ]. Some
techniques make use relevance to color histogram comparison
or motion tracking.
This paper investigates the method known as Normalized
Information Distance or (NID). NID approximates the value of
a theoretical distance between frames using the theoretical
Kolmogorov complexity. Kolmogorov complexity is a noncomputable function that can be used to measure the similarity
between objects, data or anything that can be encoded as finite
II. RELATED WORK
A great deal of researches has been done on content analysis
and segmentation of video using different techniques. In this
paper, we have based ourselves on a pervious work we have
experimented with using different segmentation techniques [a,
b] and a work done by other researchers such as:
 Zhang, Kankanhalli, and Smoliar used a pixel-based
difference method, which is one of the easiest ways to
detect if two adjacent frames are significantly
different [2,3].
 Kasturi and Jain used a statistical methods expand on
the idea of pixel differences by breaking the images
into regions and comparing statistical measures of the
pixels in those regions [4].

Zabih et al compared the number and position of
edges in successive video frames, allowing for global
camera motion by aligning edges between frames
[10].
1
III. OUR TECHNIQUE
We detect shot boundaries by calculating the distance
calculation between the two images in the types of videos, by
the result of calculations we derive how many images and
scenes per type of video with detect the value of d(x,y) when
the image or scene vary from the next image or scene , also
we compared the real calculation with the output from the
distance calculation (we discussed the result in result section)
, the distance equation is :
d(x, y)
max{(| c(xy) | -| c(y)|), ((| c(yx) | -| c(x) |)}
max{| c(x) |, | c(y)|}
are commonly used in the field of information retrieval .Recall
is defined as the proportion of shot boundaries correctly
identified by the system to the total number of shot boundaries
present (correct and missed ).
Precision is the proportion of correct shot boundaries
identified by the system to the total number of shot boundaries
identified by the system. We express recall and precision as:
Recall = Correct Scene Changes
Correct + Missed Scene Changes
Precision = Correct Scene Changes
Correct + False Scene Changes
Where :
| c(xy) | : length of compressed size of concatenated
IV. TEST BED
Table 2. Experiment results Recall and Precision
Video Type
Recall
Precision
News
93.3
78.6
Movies
95.3
100
Cookery program
90
100
Sports
88.9
72.7
Documentary
93.8
85
Comedy/ Drama
100
100
Commercials
100
91.7
To evaluate the results of our research, a database of various
compressed video types was used. The video clips were
digitized in Bmp format (total of 25442 frames) and
resolution of 320*240 pixels.
Video Type
No.of frames
News
4314
Movies
4194
Cookery program
3060
Sports
3045
Documentary
5752
Comedy/ Drama
2249
Commercials
2828
Total
25442
Table 1.Video types used & number of frames in each
EXPERIMENTAL RESULT
Results( Recall & Precision)
120
100
80
60
40
20
0
Ne
ws
The following table shows the video types used and the
number of frames were analyses
Sp
or
Do
ts
cu
m
en
Co
ta
m
ry
ed
y/
Dr
am
Co
a
m
m
er
cia
ls
images y and x .
| c(x)| : length of compressed size of image x .
| c(y)| : length of compressed size of image y .
Ideally, both recall and precision should equal 1. This would
indicate that we have identified all existing shot boundaries
correctly, without identifying any false boundaries.
Now we will describe the results obtained from this
experiment, the recall and precision for each video type as in
table 2 and figure 1 below :
M
Co
ov
ok
ie
s
er
y
pr
og
ra
m
images x and y .
| c(yx) | : length of compressed size of concatenated
Recall
Precision
In reporting our experimental results, we use recall and
precision to evaluate system performance. Recall and precision
2
Figure 5. Results obtained from experiment
(Recall&Precision) .
[5] X.U. Cabedo and S. K. Bhattacharjee: Shot detection
tools in digital video, in Proceedings Non-linear
Model Based Image Analysis 1998, Springer Verlag,
pp 121-126, Glasgow, July 1998.
[6]
V. CONCLUSION
Video segmentation has been gaining a great attention since
the world has been relaying on the internet for fast accessing
of digital multimedia files such as digital videos. Video
segmentation is the basic block for video indexing, searching,
and compression. We have presented a novel video
segmentation algorithm and demonstrated its performance.
Better features based on dual comparison measurements
further improved the segmentation.
Swanberg, D., Shu C.F., and Jain, R., “Knowledge
Guided Parsing and Retrieval in Video Databases”, in
Storage and Retrieval for Image and Video
Databases, Wayne Niblack, Editor, Proc. SPIE 1908,
February, 1993, pp. 173-187.
[7] Little, T.D.C, Ahanger, G., Folz, R.J., Gibbon, J.F.,
Reeve, F.W., Schelleng, D.H., and Venkatesh, D., “A
Digital On-Demand Video Service Supporting
Content-Based Queries”, Proc. ACM Multimedia 93,
Anaheim, CA, August, 1993, pp.427-436.
[8] R. Zabih, J. Miller, and K. Mai, A feature –based
algorithm for detecting and classifying scene breaks,
in Proceedings ACM Multimedia 95 , pages 189-200
, November 1993.
[9] J. Canny , A computational approach to edge
detection ,in IEEE Transaction on Pattern Analysis
and Machine Intelligence 8(6),pages 679-698,1986.
[10] Y. Gong, C. H. Chuan, and G. Xiaoyi, Image
indexing and retrieval based on color histograms, in
Multimedia Tools and Applications, volume 2, pages
133-156,1996.
REFERENCES
[1] H. J. Zhang, A. Kankanhalli and S. W. Smoliar,
Automatic partitioning of full-motion video , in
Multimedia Systems, volume 1 , pages 10-28, 1993.
[2] R. Kasturi, and R. Jain, Dynamic Vision, in Computer
Vision: Principles, R. Kasturi, and R. Jain, Editors,
IEEE, Computer Society Press, Washington, 1991.
[3] H. Ueda, T. Miyatake, and S. Yoshizawa, “IMPACT:
An Interactive Natural-motion-picture dedicated
Multimedia Authoring System, “in proceedings of
CHI, 1991(New Orleans, Louisiana, Apr-May, 1991)
ACM, New York, 1991, pp. 343-350.
[11] J. Meng, Y. Juan, S.-F. Chang , Scene change
detection in an MPEG compressed video sequence, in
IS&T/SPIE Symposium Proceedings , volume 2419,
February 1995.
[12] M. Tague, The pragmatics Information Retrieval
experimentation,
in
Information
Retrieval
Experiment, Karn Sparck Jones Ed., Buttersworth,
pages 59-102, 1981.
[13] Arman, F., Hsu, A., and Chiu, M-Y., “Image
Processing on Encoded Video Sequences”,
Multimedia Systems (1994) Vol.1, No. 5, pp. 211219.
[a] K. Kaabneh and H. Al-Bdour. “A New Segmentation
Technique and Evaluation.” Mu’tah Lil-Buhuth wadDirasat Journal, 19 (2): 39-51, Jan. 2004.
[4] A. Nagasaka and Y. Tanaka, Automatic video
indexing and full –video search for object
appearances , in Visual Database Systems II ,
Elsevier Science Publishers , pages 113-117, 1992.
3
[b] K. Kaabneh, O. Alia, A. Suleiman, A. Aburabela, “Video
Segmentation via Dual Shot Boundary Detection (DSBD)”,
In the proceedings of the 2nd IEEE International Conference
On Information & Communication Technologies (ICTTA’06),
April 24 - 28, 2006.
4