Medical Image Processing - Medical Image Processing: Advanced Fuzzy Set Theoretic Techniques - page 28

2.3 Image Fusion
Image fusion is a process of combining information of two or three images
to form a single image or combine some of their features in a single image.
It is used to improve the imaging quality and to reduce the randomness and
redundancy so that the assessment of medical problems becomes easier.
Its aim is to combine redundant information from multiple images to cre-
ate a fused image. The new image so generated contains a more accurate
description of the image than a source image and is more suitable for human
visual or further image processing and analysis tasks. For medical images,
fusion can lead to additional clinical information that may not be apparent
in the separate images and can also reduce storage cost by storing the single
fused image instead of multi-source images. Image fusion methods can be
grouped into three categories: pixel level, feature level and decision level. In
the pixel level, the simplest way is to take the average of two images pixel by
pixel, but many undesirable effects are observed. Many other techniques are
available such as weighted average, principal component analysis (PCA) and
Brovey transform [18,22]. In the feature level, the features involved are edges,
regions, shape, size, length or image segments. These features are then com-
bined with the similar features present in other input images to form the
final fused image.
2.4 Image Retrieval
In medical image retrieval [5,15], the number of digitally produced images
is rising enormously. Thus, more efficient image retrieval methods are
required for better management of medical information system. There are
two methods that the medical images are retrieved; these are text-based and
content-based methods [8] or a combination of the two. In the text-based
retrieval system, images are retrieved by manually annotated text descrip-
tion and traditional database techniques to manage the images. It works fast
and is reliable when the images are well annotated, but it does not work bet-
ter on un-annotated image database. Also, the annotation procedure is time
consuming and commonly results in irrelevant images. In content-based
retrieval, images are retrieved and indexed based on their visual features
such as colour, texture and shape. Commonly used features are colour fea-
tures such as red, green and blue (RGB); hue, saturation and value (HSV);
CIELab; and Luv. But as most of the medical images are in greyscale, colour
features are not used in medical image retrieval. Textural features mean
the spatial organization of pixels in the images and find the characteristics
of the image in a certain direction. Features can be obtained using Fourier