Information Technology Reference
In-Depth Information
represent image properties such as color, texture, and shape (Hassanien & Ali, 2004; Viitaniemi &
Laaksonen, 2006; Ma & Manjunath, 1999, Hassanien & Ajith 2008). In this approach, it is possible to
retrieve images similar to one chosen by the user (i.e., query-by-example). One of the main advantages
of this approach is the possibility of an automatic retrieval process, contrasting to the effort needed to
annotate images.
The work introduced in this chapter is based on the second retrieval approach. Image similarity is
typically defined using a metric on a feature space. Numerous similarity metrics have been proposed
so far. The search results are combined with existing textual information and collections of other fea-
tures via intelligent decision support systems. In this paper, we use a new similarity function based on
the rough set theory (Grzymala-Busse, Pawlak, Slowinski, & Ziarko, 1999; Kent, 1994; Pawlak, 1991;
Pawlak, 1982; Pawlak, Grzymala-Busse, Slowinski, & Ziarko, 1995, Jafar M. Ali, 2007). This theory
has become very popular among scientists around the world. Rough sets data analysis was used for
the discovery of data dependencies, data reduction, approximate set classification, and rule induction
from databases. The generated rules represent the underlying Semantic content of the images in the
database. A classification mechanism is developed by which the images are classified according to the
generated rules.
r ela ted work and Pro Ble M definition
Image classification and retrieval methods aim to classify and retrieve relevant images from an image
database that are similar to the query image. The ability to effectively retrieve nonalphanumeric data is
a complex issue (Jafar M. Ali, 2007). The problem becomes even more difficult due to the high dimen-
sion of the variable space associated with the images. Image classification is a very active and promising
research domain in the area of image management and retrieval. A representative example is presented
by (Lienhart & Hartmann, 2002)) who implemented and evaluated a system that performs a two-stage
classification of images: first, photo-like images are distinguished from nonphotographic ones, followed
by a second round in which actual photos are separated from artificial, photo-like images, and nonpho-
tographic images are differentiated into presentation slides, scientific posters, and comics. This scheme
is neither exclusive nor exhaustive; many images fall into multiple categories. Some systems have used
edge and shape information that is either supplied by the user or extracted from training samples (Saber
& Tekalp, 1998). However, such systems require detailed region segmentation. Segmentation has been
used to extract region-based descriptions of an image by NeTra, Blobworld, and SIMPLIcity (Ma &
Manjunath, 1999; Carson, Thomas, Belongie, Hellerstein, & Malik, 1999; Wang, Li, & Wiederhold,
2001). NeTra and Blobworld present a user with the segmented regions of an image. The user selects
regions to be matched, together with attributes such as color and texture. SIMPLIcity is able to match all
segmented regions automatically. However, a user's Semantic understanding of an image is at a higher
level than the region representation. Often it is difficult for a user to select a representative region for
the entire image; coupled with the inaccuracy of automatic segmentation, the retrieved results do not
match the user's intuition, or understanding of the images. An object is typically composed of multiple
segments with varying color and texture patterns. One or more segmented regions are usually not suf-
ficient to address Semantic object representation.
A key feature of our approach is that segmentation and detailed object representation are not required.
Our approach is a texture-color-based image retrieval system using a similarity approach on the basis
Search WWH ::




Custom Search