Database Reference
In-Depth Information
Mining Image Databases by Content
Gerald Schaefer
Department of Computer Science, Loughborough University, Loughborough, U.K.
Abstract.
Visual information is becoming more important and at a
rapid rate. However, creators and users are reluctant to annotate visual
content making it dicult to search these collections. Content-based im-
age retrieval (CBIR) techniques extract visual descriptors directly from
image data and can hence be used in situations where textual informa-
tion is not available. In this paper, we give a brief introduction on some
of the basic colour descriptors that are employed in CBIR.
1
Introduction
While personal image collections may contains thousands of images, commercial
image repositories can comprise several million images [1]. Effective and ecient
methods for querying these collections are highly sought after. While images
are rarely annotated [2], content-based image retrieval (CBIR) techniques [3],
which extract image features describing colour, texture, shape etc. attributes to
formulate a query, can be successfully employed. In this paper, we give a brief
introduction on some of the basic colour descriptors that are used for CBIR.
2 Content-Based Image Retrieval by Colour
Colour has been shown to be one of the most effective feature types for CBIR.
The simplest colour descriptor for CBIR are colour moments [4] which are de-
fined by central normalised moments of the colour distribution of an image
(usually mean, standard deviation and kurtosis in each colour channel). Visual
(dis)similarity between two images can be described by the
L
1
norm between
their moment vectors.
Swain and Ballard [5] introduced the use of colour histograms, which record
the frequencies of colours in the image, to describe images in order to perform
object recognition and image retrieval. As similarity measure they introduced
histogram intersection which quantifies the overlap between two histograms and
can be shown to be equivalent to an
L
1
norm.
Rather than using colour histograms, a more compact descriptor for encoding
the colour distribution of images is a colour signature. Colour signatures are a
set
where
c
i
define colour co-ordinates and
ω
i
their associated weights (i.e., their frequencies in the image). A common way of
deriving colour signatures for images is through a clustering process. Once colour
signatures for images are determined, these signatures can be compared by a
(
c
1
,ω
1
)
,
(
c
2
,ω
2
)
,...,
(
c
m
,ω
m
)
{
}
