ISO/IEC JTC 1/SC 34 N148
Date: 2000-06-10
Title:Korean Comments on
JTC1/SC34 N0066 (A method for multimedia context based image retrieval using color moment in
wavelet transformation area )
Source: Ki-Joung
Kang, Multimedia Technical Lab,
Korea Telecom
Project:
Status: To
be reviewed and discussed at Regular meeting, June 2000
1. Summary
In
Granada (Spain) meeting, 19 ~ 23 April 1999, a contribution of “Multimedia
Retrieval System Structure and Image Retrieval (N0066)”was presented. In this contribution, a method for
multimedia context based image retrieval using color moment in wavelet
transformation area is proposed as an explanation of the contribution, N0066.
2. Introduction
XML
(Extensible Markup Language) was proposed as a standardized web document format
at W3C (World Wide Web Consortium) in 1996. A standardized text format, XML, is
a language for next generation internet. It resolves deficiencies of HTML
(Hyper-Text Markup Language) and takes advantages of HTML and SGML (Standard
Generalized Markup Language). XML makes a structured document transmitted
through web and gives a method to describe user-defined document. Web
applications can process XML.
W3C
approved XML as a standard to make up for the weak points of HTML. XML can be
considered as internet version of SGML. W3C launched SMIL (Synchronized
Multimedia Integration Language) WG(Working Group). SMIL is used to integrate
multimedia files stored at XML based server and to present the file at the
execution point. In SC34 WG3, a standard for interactive multimedia document,
ISMID, is being developed. In this contribution, a method for multimedia
context based image retrieval using color moment in wavelet transformation area
is proposed as an application of these standards.
3. Conventional Method
In the rapid-changing environment of communication network and
multimedia technology, the study on storage, administration and retrieval for
multimedia data is a primary issue.
Wavelet transformation method is used to extract
characteristic vector [1]. After characteristic vector for each channel of R,
G, and B is configured, the query is performed. Wang uses wavelet coefficient
of level 4 in order to configure characteristic vector after wavelet
transformation is performed with the help of Daubeche-8 wavelet. Direct
comparison between wavelet coefficients is used to measure the similarity of
images. These methods are useful in the case that overall color distribution is
adopted to measure the similarity between images. But they are incongruent in
the case that object shape is used for the similarity measurement. As the
dimension of characteristic vector is increased, the time required to retrieve
an image is extended.
4. Proposed Method
In this contribution, a context based image retrieval method
using color and moment in the wavelet transformation area is proposed. In the
query using the color of image, energy of wavelet coefficient in the HSV color
space that is fit for the characteristic of human sense of sight. In the query
using the moment, wavelet maxima of high frequency band in the wavelet
transformation area is used. So query method is independent to the change of
movement, rotation and size of the object. The size of characteristic vector is
reduced to make the query time shorten. System configuration of the proposed
method is shown in the following figure.
Fig 1. System Configuration
1) Extraction of characteristic vector
After the color space of query image is transformed to the HSV
color space, 2D DWT is performed for each channel. In the configuration of
characteristic vector, wavelet coefficient was not directly used in order to
reduce the size of characteristic vector. The value of energy is used instead
as Albuz was [3]. In the case of ordinary image, the objects in the image are
usually located in the center of the image. Each band of wavelet transform area
was divided in 5 fixed areas as a result of consideration of camera shutter
motion. Fig 2 shows the divided areas. Energy of wavelet coefficient for each
area was calculated for each HSV channel. At this time, the size of
characteristic vector is 5(number of area) x 3 (number of channel) x 4byte =
60byte.
For the query using moment, 2nd central moment
value for each HSV channel using wavelet maxima in the three high frequency
band(LH, HL, HH) of wavelet transformation area is used as a characteristic
vector.
Fig 2. Area Configuration
2) Measurement of similarity
Characteristic vector of query image and one of images stored
in a database is compared to measure similarity between images[4]. Euclidean
distance is used in the measurement of similarity as following formula 1:
(formula 1)
In the user interface, there are three query methods of ?r)color’, ?r)moment’
and ?r)color and moment’[5]. In the query using color, the weight of V value is
relatively low comparing to the value of hue and saturation. It reduces the
effect of non-uniformity of light. In the proposed method, weight of hue and saturation
is 2 and weight of value is 1. The query using moment is performed to compare
wavelet maxima moment by the Euclidean distance of formula 1 in the high
frequency band of each channel of H, S, and V.
5. Conclusion
In the context based image retrieval using characteristic vector, the size of characteristic vector should be reduced as far as the retrieval efficiency is not rapidly declined.
In this
contribution, wavelet coefficient was not directly used as a characteristic
vector in the wavelet transformation area. Energy of coefficient of each area
is used to reduce the size of characteristic vector and to increase the
efficiency of retrieval.
6. Reference
1. Charles E.
Jacobs, 'Fast Multiresolution Image Query', Proceedings of the 1995 ACM SIGGRAPH,
New
2. James Ze Wang, 'Content-based Image
Indexing and Searching using Daubechies' Wavelet',Journal of Digital Library,
1998
3. Elif Albuz, 'Scalable Image Indexing and Retrieval using Wavelets', Technical Report, University of Delaware, 1998,11
4. Charles E. Jacobs, 'Fast Multiresolution Image Query', Proceedings of
the 1995 ACM SIGGRAPH, New York, 1995
5. Y.T.Chan, 'Wavelet Basics', Kluwer Academic Plublishers,1995