Image Processing Reference
In-Depth Information
is proportional to the
brightness
of the corresponding point in the scene; its value is often
derived from the output of an A/D converter. The matrix of pixels, the image, is usually
square and we shall describe an image as
N
×
N
m
-bit pixels where
N
is the number of
points along the axes and
m
controls the number of brightness values. Using
m
bits gives
a range of 2
m
values, ranging from 0 to 2
m
- 1. If
m
is 8 this gives brightness levels ranging
between 0 and 255, which are usually displayed as black and white, respectively, with
shades of grey in between, as they are for the
greyscale image
of a walking man in Figure
2.1
(a). Smaller values of
m
give fewer available levels reducing the available contrast in an
image.
The ideal value of
m
is actually related to the signal to noise ratio (bandwidth) of the
camera. This is stated as approximately 45 dB and since there are 6 dB per bit, then 8 bits
will cover the available range. Choosing 8-bit pixels has further advantages in that it is very
convenient to store pixel values as
bytes
, and 8-bit A/D converters are cheaper than those
with a higher resolution. For these reasons images are nearly always stored as 8-bit bytes,
though some applications use a different range. The relative influence of the 8 bits is shown
in the image of the walking subject in Figure
2.1
. Here, the least significant bit, bit 0
(Figure
2.1
(b)), carries the least information (it changes most rapidly). As the order of the
bits increases, they change less rapidly and carry more information. The most information
is carried by the most significant bit, bit 7 (Figure
2.1
(i)). Clearly, the fact that there is a
walker in the original image can be recognised much better from the high order bits, much
more reliably than it can from the other bits (notice too the odd effects in the bits which
would appear to come from lighting at the top left corner).
Colour images
follow a similar storage strategy to specify pixels' intensities. However,
instead of using just one image plane, colour images are represented by three intensity
components. These components generally correspond to red, green, and blue (the RGB
model) although there are other colour schemes. For example, the CMYK colour model is
defined by the components cyan, magenta, yellow and black. In any colour mode, the
pixel's colour can be specified in two main ways. First, you can associate an integer value,
with each pixel, that can be used as an index to a table that stores the intensity of each
colour component. The index is used to recover the actual colour from the table when the
pixel is going to be displayed, or processed. In this scheme, the table is known as the
image's
palette
and the display is said to be performed by
colour mapping
. The main
reason for using this colour representation is to reduce memory requirements. That is, we
only store a single image plane (i.e. the indices) and the palette. This is less than storing
the red, green and blue components separately and so makes the hardware cheaper and it
can have other advantages, for example when the image is transmitted. The main disadvantage
is that the quality of the image is reduced since only a reduced collection of colours is
actually used. An alternative to represent colour is to use several image planes to store the
colour components of each pixel. This scheme is known as
true colour
and it represents an
image more accurately, essentially by considering more colours. The most common format
uses 8 bits for each of the three RGB components. These images are known as
24-bit
true
colour and they can contain 16 777 216 different colours simultaneously. In spite of requiring
significantly more memory, the image quality and the continuing reduction in cost of
computer memory make this format a good alternative, even for storing the image frames
from a video sequence. Of course, a good compression algorithm is always helpful in these
cases, particularly if images need to be transmitted on a network. Here we will consider the
processing of grey level images only since they contain enough information to perform
feature extraction and image analysis. Should the image be originally colour, we will
