Page 288 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 10 Deep neural network in medical image processing 279

Enhancement: It is the process of modifying or adjusting dig-
ital images for further analysis or display. Most common methods
of enhancement are as follows:
• Tweaking brightness
• Adjusting contrast of the images
• Noise removal
• Sharpen the image
• Histogram modeling
• Image averaging
• Using filters (linear, nonlinear)
• Zooming
Restoration: Image restoration is also an important step in
image processing. In this step, we intend to reverse the compen-
sate or undo the effects of degradation. Various types of methods
for image restoration are as follows:
• Inverse filter
• Weiner filter
• Wavelet restoration
• Blind deconvolution
Segmentation: It is the process of dividing a digital image into
multiple segments each containing pixels with similar attributes.
Image segmentation is the first step to divide the image into
multiple images based on various factors, for example, different
objects in the picture or change of scenery. The successful analysis
of an image greatly depends on the reliability of segmentation, but
in practice, it is a very challenging problem. Segmentation algo-
rithms can broadly be categorized in two different categories.
Contextual: Contextual Segmentation exploits spatial relation-
ship between features in an image e.g grouping pixels with
similar grey levels together, this type of algorithms are particu-
larly successful in separating individual objects as it accounts
for proximity of pixels that belongs to a single object.
Noncontextual: Noncontextual segmentation techniques basi-
cally group together pixel based on a common attribute like color
or gray level; it does not take into account the spatial relationship
between different objects in the image.
Description/identification: This is a step which is followed by
segmentation. When the image is partitioned into different well-
defined regions/segments with segmentation algorithms, then
those segments are matched against various well-known param-
eters to classify them into various classes such as differentiating
between shapes or land and water or land and sky.
Object recognition: This step is a further enhancement to
description/identification as it tries to further classify the
broader identified segments into finer classes and in turn
convert the image to a plethora of meaningful tags, for example,