16
L.-K. SOH, C. TSATSOULIS, AND B. HOLT
Opening generally trims off peninsulas, breaks bridges, and solidifies channels. Closing generally closes inlets, absorbs islands, and fills lakes. Two examples of closing and
opening are shown in Fig. 3. Given the core and skin images, we perform cleaning on
each before executing the restricted growing algorithm that grows the core objects
within their corresponding skin objects.
Figure 4 shows a series of images depicting different stages of the execution of the
restricted growing concept. Figure 4a shows a portion of a STAR-2 SAR sea ice imagery.
The image consists of bright ice floes and dark leads. Figure 4b shows the segmented
image of Fig. 4a. The image segmentation technique used was binary global thresholding. Note that the ice floes are connected. These networks are inseparable under conventional segmentation. Figures 4C and 4d show the core and skin images, respectiveFig. 3. a Original image of a
peninsula. b-e Closing (dilation
and erosion) and opening (erosion and dilation). f Original
image of an inlet. g-j closing
(dilation and erosion) and
opening (erosion and dilation).
Note that the peninsula and the
inlet are eliminated while the
shape of the object region is
still preserved. The operator
window is 3 x 3. Dark pixels
denote object pixels; unshaded
pixels denote nonobject pixels
Fig.4a-h. Execution of the
restricted growing concept.
a Portion of an original STAR-2
SAR image (May 1987), b segmented image, c core image,
d skin image, e cleaned core
image, f cleaned skin image,
g final grown image output,
h superimposition of the floe
boundaries on the original
image
(I)
(g)
(h)
(i)
(j)
L.-K. SOH, C. TSATSOULIS, AND B. HOLT
Opening generally trims off peninsulas, breaks bridges, and solidifies channels. Closing generally closes inlets, absorbs islands, and fills lakes. Two examples of closing and
opening are shown in Fig. 3. Given the core and skin images, we perform cleaning on
each before executing the restricted growing algorithm that grows the core objects
within their corresponding skin objects.
Figure 4 shows a series of images depicting different stages of the execution of the
restricted growing concept. Figure 4a shows a portion of a STAR-2 SAR sea ice imagery.
The image consists of bright ice floes and dark leads. Figure 4b shows the segmented
image of Fig. 4a. The image segmentation technique used was binary global thresholding. Note that the ice floes are connected. These networks are inseparable under conventional segmentation. Figures 4C and 4d show the core and skin images, respectiveFig. 3. a Original image of a
peninsula. b-e Closing (dilation
and erosion) and opening (erosion and dilation). f Original
image of an inlet. g-j closing
(dilation and erosion) and
opening (erosion and dilation).
Note that the peninsula and the
inlet are eliminated while the
shape of the object region is
still preserved. The operator
window is 3 x 3. Dark pixels
denote object pixels; unshaded
pixels denote nonobject pixels
Fig.4a-h. Execution of the
restricted growing concept.
a Portion of an original STAR-2
SAR image (May 1987), b segmented image, c core image,
d skin image, e cleaned core
image, f cleaned skin image,
g final grown image output,
h superimposition of the floe
boundaries on the original
image
(I)
(g)
(h)
(i)
(j)