Medical Image Processing - Medical Image Processing: Advanced Fuzzy Set Theoretic Techniques - page 29

transform, Gabor filters, wavelets, co-occurrence matrices and many others.
In medical images, textural features are required as they reflect the details
of the image structure. Shape features include edges, boundaries, contour,
curve and surfaces.
Content-based image retrieval is a system for browsing, searching and
retrieving images from a large database. It is a problem of searching for
digital images from the large database where images are retrieved based
on image features. In this retrieval system, a query image is required and
is allowed to match the images in the database based on some similarity
measures. Image features that may be colour or texture or shape for both
the query image and the images in the database are used in matching. In the
matching procedure, either the entire image is used or subimages are used
depending on the user's choice. For each image in the database, image fea-
tures are computed and likewise features of the query image are also com-
puted. Using the similarity measure, a match is found between the query
image and the images in the database.
Management and access of these medical images become very complex.
Access is mainly based on patient identification or study characteristics. The
purpose of a medical image retrieval system is to deliver the needed informa-
tion at the right time and to right person to improve the quality of the care
process. In the decision-making process, the retrieval system will find other
images of the same modality and same anatomical region of the same disease.
Storing and accessing these large numbers of images are very important. As
it requires large space, data reduction techniques are used. For feature reduc-
tion, PCA is used, which is also called Karhunen-Loeve transform (KLT).
2.5 Fuzzy Processing of Medical Images
Medical image processing may be done using crisp or fuzzy set theory.
Medical image processing is intended as a central resource for information
of image processing in the medical field. Classical methods based on crisp
set theory are mentioned in many texts. In crisp set, the structures present
in the image are considered to have a rigid demarcating boundary and crisp
methods consider bivalent logic, and the degree of belongingness (member-
ship degree) of a pixel present in an image is either 0 or 1. But the objects
in the real world do not always have a rigid demarcating boundary and so
there is a gradual transition of the membership degree from zero to unity
membership, and this is what fuzzy set theory deals with as introduced by
Zadeh [24]. It considers one uncertainty, which is the degree of belonging-
ness of an element in a set. The application of fuzzy set theoretic concepts in
image processing took formal shape only in the 1980s with the pioneering
research carried out by Pal et al. [16,17] and Rosenfeld and Pal [20,21].