Computer Vision Analysis of a Melting Interface Problem with Natural Convection - Visual Computing: Scientific Visualization and Imaging Systems - page 216

Fig. 12.6 Picture of the

experiment

Fig. 12.7 Example of an
image representation

Image segmentation is the process of partitioning an image into meaningful
parts [ 2 ]. Segmentation techniques can be thresholding, boundary detection, or
region growing.
In the thresholding process, generally it used the analysis of pixels grayscale
frequency (histogram) to determine a threshold (l) parameter (Eq. 12.13 ). The
image result is a new one I(x,y)[ 20 ].

I ð x ; y Þ ¼ 0 ;

f ð x ; y Þ \l

ð 12 : 13 Þ

255 ; f ð x ; y Þ l

This method can achieve good results in images with bimodal histograms
(Fig. 12.8 ), different from the example image histogram shown in Fig. 12.9 , where
there is no two noticeable picks. In bimodal histograms case, it is simple to choice
a threshold parameter l to segment the two different regions in an image. However,
the method of amplitude threshold is not a good procedure in many cases, as can
be seen in Fig. 12.9 . The reason of this bad result is the fact that the luminance
isn't a distinguishable feature for water and ice in this particular example.