DataVS1
Series
Instruction
Manual
80
14.5.
Binarization
This is a process that converts any given image to a two-level image (in our instance, black and
white). Conversion is based on a threshold that classifies the individual pixels as "level 255" (or
"white") or "level 0" (or "black") depending on whether they are above or below set threshold.
Let us consider a shot of a dark object on a bright background. Binarisation is expressed as follows:
S
j
i
I
if
S
j
i
I
if
j
i
B
)
,
(
255
)
,
(
0
)
,
(
In other words, the pixel with coordinates (i,j) of the source image I will be coloured black in the
binarised image B if its (brightness) value is found to be below the threshold S, or will be coloured
white if its value is above the threshold.
The purpose of this operation is to highlight just the key features of the image to minimise the
computational load associated with its analysis.
The critical issue in this process is how to select the threshold S.
Let us consider a histogram of a grey scale image (representing the number of pixels as a function of
brightness value)
number of pixels
brightness value: 0 – 255
grey scale image
image histogram
As you can see, the brightness values of image pixels are clustered around two values ("bimodal"
histogram): a low value that represents dark points (the object) and a high value that represents bright
points (the background).
The most effective value is found mid-way between the two (128). This will turn out a binarised image
where the object is neatly and accurately represented. With a lower threshold (90), part of the object
would be treated as background; a higher threshold (175) would cause part of the background to be
classified as object
Image binarised with S = 128
Image binarised with S = 90
Image binarised with S = 175
Obviously enough, an image where object and background are not clearly differentiated (similar
brightness values) will turn out a histogram with all pixels clustered around one value, and setting the
appropriate threshold will prove quite difficult.
