Computer Vision Analysis of a Melting Interface Problem with Natural Convection - Visual Computing: Scientific Visualization and Imaging Systems - page 222

method uses two 3 9 3 masks that approximate the derivatives of the image, for
direction x and direction y.

2
4

3
5

2
4

3
5

101
202
101

121
000
1 2 1

ð 12 : 18 Þ

Gx Gy

Using the obtained results from both directions, the gradient of the image can
be calculated (Eq. 12.19 ).

G ¼

q

G x þ G y ð 12 : 19 Þ

The next step is to determine the edge directions in the image. Using the values
of the gradient and Eq. 12.20 , we can determine these directions.

G y
G x

h ¼ a tan ð 12 : 20 Þ

When the algorithm is implemented we must consider the cases where Gx is
equal to 0. In this case, we will generate a mathematical error. Therefore, in these
points the algorithm must analyze the gradient in Gy direction in order to decide
which is the edge direction (0,45 ,90 , 135).
Then the algorithm realizes a suppression process. This process consists in
analyze, using a 3 9 3 matrix, all the points in all obtained directions calculating
their gradient angle. We will consider as an edge only the points that have higher
grayscale intensity than their correspondent neighbors in the opposite directions.
Finally, to find the edges, the image is analyzed using two threshold points, a
high and a low one. When a pixel is greater than the high threshold value, its
neighbor pixels need to have higher values than the low threshold point, in order to
generate a continuous edge.

12.3.3 Image Processing Results

The Canny method was applied in five images of the experiment presented in Sect.
2.2 . To reduce the possibilities of error in this algorithm and to obtain more
efficient computational results, the images had been divided into two parts,
because in this experiment the space occupied by the ice wasn't higher than 50 %
of the total reservoir area (Fig. 12.14 ).
The results of Canny method in the images are shown in Fig. 12.15 .
Applying the algorithm, it is possible to separate completely water and ice in
each image. These images are used as masks in the original images (Fig. 12.16 ).