2 Identifying Ice Floes and Computing Ice Floe Distributions in SAR Images
13
ness whereas in restricted growing, an object is grown while maintaining its separation
from other objects. Our algorithm consists of five sequential tests as follows:
Test 1: Potential Growing Test. If a pixel in the core image is a nonobject pixel and its
corresponding pixel in the skin image is an object pixel, then the pixel is qualified for
further tests.
This test selects only nonobject pixels for potential growth. Note that pixels that are
nonobject in both core and skin images are deemed as true nonobject pixels without
any potential for becoming object pixels.
Test 2: Isolation Test. If the pixel in the core image does not have an object pixel in
its core image 8-neighborhood, then the pixel is disqualified from the growing process.
The isolation test avoids erroneous separation within an object. For example, small
dark specks in an object would be eroded to nonobject pixels during the generation of
core image. If allowed to grow, these non object holes could become individual object
regions within the object that enclosed them. With this test, however, the nonobject holes
would be consumed by the growth of the encroaching core objects.
Test 3: Connectedness Test 1. If the pixel in the core image has seven or eight object
pixels in its core image 8-neighborhood, then the pixel is grown.
If the 8-neighborhood of a pixel has seven or more object pixels, that means all object
pixels in that neighborhood are connected. Hence, the growth of that pixel from nonobject to object does not damage the existing (or nonexisting) separation.
Test 4: Connectedness Test 2. If the pixel in the core image has less than two "role
transitions" in its core image 8-neighborhood, then the pixel is grown.
A role transition is a switch from object to nonobject while following the neighborhood in the sequence shown in Fig. 2a. If less than two role transitions are found, then
all the object pixels are connected and further growth does not damage the separation.
Test 5: Connectedness Test 3. If the pixel in the core image has its core image 8-neighborhood matching one of the four corner patterns, then the pixel is grown.
The corner patterns are shown in Fig. 2b. Any of these patterns could have two or
more role transitions yet have all its object pixels connected, qualifying the pixel for
the growing process.
The restricted growing algorithm performs these five tests sequentially. At each iteration, the algorithm moves in a raster manner, top to bottom and left to right, until the
whole image is scanned. Any growing (a pixel from nonobject to object) is implemented immediately after each pixel has been inspected. To delay the growth effect on the
Fig. 2. a The sequence followed
by the connectedness test 2 of
the restricted growing algorithm. The center pixel is the
current pixel. The sequence
starts at pixel number 1, then 2,
... , and ends at pixel number 8.
b Four corner patterns examined during the connectedness
test 3 of the growing algorithm.
Dark pixels denote object pixels; unshaded pixels denote
"don't care" pixels
8
7
6
1
0
5
(a)
2
3
4
(b)
13
ness whereas in restricted growing, an object is grown while maintaining its separation
from other objects. Our algorithm consists of five sequential tests as follows:
Test 1: Potential Growing Test. If a pixel in the core image is a nonobject pixel and its
corresponding pixel in the skin image is an object pixel, then the pixel is qualified for
further tests.
This test selects only nonobject pixels for potential growth. Note that pixels that are
nonobject in both core and skin images are deemed as true nonobject pixels without
any potential for becoming object pixels.
Test 2: Isolation Test. If the pixel in the core image does not have an object pixel in
its core image 8-neighborhood, then the pixel is disqualified from the growing process.
The isolation test avoids erroneous separation within an object. For example, small
dark specks in an object would be eroded to nonobject pixels during the generation of
core image. If allowed to grow, these non object holes could become individual object
regions within the object that enclosed them. With this test, however, the nonobject holes
would be consumed by the growth of the encroaching core objects.
Test 3: Connectedness Test 1. If the pixel in the core image has seven or eight object
pixels in its core image 8-neighborhood, then the pixel is grown.
If the 8-neighborhood of a pixel has seven or more object pixels, that means all object
pixels in that neighborhood are connected. Hence, the growth of that pixel from nonobject to object does not damage the existing (or nonexisting) separation.
Test 4: Connectedness Test 2. If the pixel in the core image has less than two "role
transitions" in its core image 8-neighborhood, then the pixel is grown.
A role transition is a switch from object to nonobject while following the neighborhood in the sequence shown in Fig. 2a. If less than two role transitions are found, then
all the object pixels are connected and further growth does not damage the separation.
Test 5: Connectedness Test 3. If the pixel in the core image has its core image 8-neighborhood matching one of the four corner patterns, then the pixel is grown.
The corner patterns are shown in Fig. 2b. Any of these patterns could have two or
more role transitions yet have all its object pixels connected, qualifying the pixel for
the growing process.
The restricted growing algorithm performs these five tests sequentially. At each iteration, the algorithm moves in a raster manner, top to bottom and left to right, until the
whole image is scanned. Any growing (a pixel from nonobject to object) is implemented immediately after each pixel has been inspected. To delay the growth effect on the
Fig. 2. a The sequence followed
by the connectedness test 2 of
the restricted growing algorithm. The center pixel is the
current pixel. The sequence
starts at pixel number 1, then 2,
... , and ends at pixel number 8.
b Four corner patterns examined during the connectedness
test 3 of the growing algorithm.
Dark pixels denote object pixels; unshaded pixels denote
"don't care" pixels
8
7
6
1
0
5
(a)
2
3
4
(b)