VIDEO DATA MINING PATTERN DISCOVERED USING
STATISTICAL MODELING APPROACH
B.OBULIRAJ
Abstract– The task to discover
useful editing patterns from professional video,
such as movie , is one of the main purpose of Video
Data Mining. But data mining an multimedia data like a movie is a challenging
task due to the complicated contents of a multimedia data particularly the
discovered patterns need to be supported by the semantic features, because the
features tell the amateur editors how to use the corresponding patterns during
the process of editing a new video. Along with this indexing mechanism a
histogram-based color descriptors also introduced to reliably capture and
represent the color properties of multiple images. Including of this a classification
approach is also carried out by the classified associations and by assigning,
each of them with a class label, and uses their appearances in the video to construct
video indices. Our experimental results demonstrate the performance of the proposed
approach.
Index Terms– video indices, video
indexing, histogram
1
INTRODUCTION
Large amounts of image data is
being stored in the databases everyday. All the data can only be used
efficiently only when the correct data is retrieved successfully.The existing
successful systems are text based and are not effective at retrieving the images.
For efficient retrieval the image properties are used to retrieve the appropriate
images which do not depend on image descriptions.. The proposed system identifies
the objects in the images based
on the colour distribution and performs the search. A
colour histogram is generated and image segmentation is done to obtain the suitable
images.
1.1 Related
Work in Video Indexing
The trend of multimedia
source digitization has accelerated the combining of many multimedia processing
techniques, such as storage, editing, and presentation on computers.
Accordingly, techniques are expected for new methods of presenting and creating
multimedia content.. However, in the retrieval of video data with motion or shapes,
the retrieval method of a user is not necessarily in agreement with the
particular data to be retrieved even when the database system provides graphic
user interface (GUI) tools. This paper describes a retrieval technique based on
gestures for video images. By using gestures, a user can input spatiotemporal
information intuitively.
Figure 1 System Architecture
2.
THE PROPOSED APPROACH
2.1 Input Video File Processing
In Video File processing applications the
histogram mechanism is proposed. Which utilize the
motion histogram for video retrieval, clustering and classification. The videos
include different videos (e.g., file downloading, Earth Movement), Folder Copy
etc. The duration of each video segment varies from 1 to 5 s and there are a total
of 50 frames. We expect various types of motion content in the different videos
and therefore, they form a suitable data set in which to test the proposed
motion
histogram.
2.2 Indexing Of Frame
For any data retrieval the indexing is treated as an
important aspect. As mentioned earlier there are different types of systems
that fetch work based on different algorithms and the retrieved
image can be the same query image or a
part of the query image or even objects in the image.
Sometimes the query can be “retrieve ten most similar images to the given
image” which can be called as the Nearest-neighbour. In the existing systems the
performance is proportional to the database size. The efficiency of the system should
depend on the number of similar images rather the total number of images in the
database.
2. 3 Histogram Search
Histogram search algorithms characterize an image by
its color distribution or histogram. A histogram is nothing but a graph that
represents all the colors and the level of their occurrence in an image irrespective
of the type of the image. Few basic properties about an image can be obtained
from using a Histogram. It can be used to set a threshold for screening the images.
The shape and the concentration of the colors in the histogram will be the same
for similar objects even though they are of different colors..
Figure 2.
Color histogram generation
Process
In general any image contains useful and
unwanted information. The system has to differentiate
between the both. Consider the below image where the person reading a book is
the useful information and the background, people and the market is the unwanted
data. The system has to group together the repeated pattern to identify the objects
in the image. For example below is given the array for the part of the shirt
and this pattern is repeated again
Figure 3 Histogram
Consider the above image where the (small part of the
person’s) shirt is enlarged and the respective representation in the form of the
array is given. The basic concept behind the histogram generation is simple.
Each pixel in the image is scanned and the respective color or intensity value
is obtained for the pixel.
iColor = (16 * p1[0]) + p1[1] * 4 + p1[2];
Then a graph is generated with total number of pixels
against the pixel intensity. An array variable is chosen to store the different
intensities and the counter increases for each repeated intensity counting the
total number of occurrences of that particular color or intensity.
iHistoArr[iColor] = iHistoArr[iColor] +1
3
EXPERIMENTAL RESULTS
Different experiments were
performed, one for assessing computational performance, and one assessing
robustness with respect to validate the methods we have described, we
implemented the components of the video frame based retrieval system and tested
with a general purpose image database including about 100 videos. The table
given below shows the time taken for Splitting Number of frames from the image database.
Table 3.1
Average Search Time
These images are stored in JPEG format
with size 384 _ 256 or 256 _ 384. The system was
written in the VB programming language and compiled on Windows platform. In
this section, we describe the training concepts and show indexing results.The
table given below shows the number of frames spitted from the videos.
Table 3.2
Average Search Time For a
particular frame
Fig. 3.1. The model input frames of
different types of test videos
4.
CONCLUSIONS
In this Research, the need for
an efficient video based indexing & Video Frame Based retrieval system is
identified and the problems with the existing systems are discussed. A new
system is proposed for different types of Video Frames and the structure of the
Frames and the way they are indexed and stored inside the database are
discussed. Different types of images can be processed using the system. The images
in the database are compared with the properties of the Video Frame based on color
histogram and appropriate results are displayed to the user. The architecture
is designed by studying different existing systems and current research areas. Research
in the fields of image processing, segmentation, edge detection, pattern recognition
and more is performed to design the system
.
5.
REFERENCES
[1] Nevenka Dimitrova, Hong-Jiang Zhang, Behzad
Shahraray, Ibrahim Sezan, Thomas Huang, and Avideh Zakhor, “Applications of
video-content analysis and retrieval,” IEEE Multimedia Magazine, vol. 9, no. 3,
pp. 42 – 55, July 2002.
[2] Asif 2005, CoderSource, 18 June 2009,
[3] Tang L 2007, University of Surrey,
18 June 2009,
[4] R. Benmokhtar and B. Huet,“Neural
network combining classifier based on Dempster-Shafer
theory for semantic indexing in video content,” International MultiMedia
Modeling Conference, vol.
4351, pp. 196–205, Singapore, 2007.
[5] D.K.
Park, Y.S. Jeon and C.S.
Won,
“Efficient use of local edge histogram descriptor,”ACM
Workshops on Multimedia, pp. 51–54, USA, 2000.
[6] M. Rautiainen and T. Seppanen,“Comparison of
visual features and fusion techniques in automatic detection of concepts from
news video,” IEEE International Conference on Multimedia & Expo, The
Netherlands, 2005.
[7] E. Allwein, R. Schapire, and Y. Singer, “Reducing
multiclass to binary : A unifying approach for margin classifiers.” Journal of Machine
Learning Research, vol. 1, pp. 113–141,
2000.
[8] Steyvers, M., Smyth, P., Rosen-Zvi, M., & Griffiths, T.(2004).
Probabilistic Author- Topic Models for Information Discovery. The Tenth ACM
SIGKDD International Conference on Knowledge Discovery and Data Mining. Seattle, Washington.
[9] Tardini, G. Grana C., Marchi, R., Cucchiara, R.,
(2005). Shot Detection and Motion Analysis for Automatic MPEG-7 Annotation of
Sports Videos. In 13th International Conference on Image Analysis
and Processing.
[10] Witten, I. and
Frank, E. (2005) "Data Mining: Practical machine learning tools and techniques",
2nd Edition, Morgan Kaufmann, San
Francisco, 2005
[11] Worring, M., Snoek, C.. (2006).
Semantic Indexing and Retrieval of Video. Tutorial at
ACM Multimedia.
No comments:
Post a Comment