18
L.-K. SOH, C. TSATSOULIS, AND B. HOLT
tering, (5) separation of open water from the remainder of the ice cover not identified
as floes, termed water/ice mixture extraction, and (6) computation of floe size distribution and coverage statistics. Discussing stage 5) further, after floe extraction and filtering, much of the ice cover is not identified as distinct floes. This primarily includes
pieces of ice that are on the order of the SAR resolution (variable but generally 100 m
or less) and new ice forms. We have termed this remaining ice water/ice mixture' for
convenience. With the knowledge of the floe, water/ice mixture, and water distributions,
we are able to compute complete coverage statistics within a SAR sea ice image. Section 2.5 below discusses an application for examining summer ice where knowledge of
the quantities of identifiable floes and their sizes, open water, and the remaining
water/ice mixture are important.
1. Image Enhancement. To ensure well-defined floes and reduction of radar speckle noise effects (Ulaby et al. 1982), we use median filtering (Huang et al. 1979). Median
filtering enhances the image by eliminating maxima or minima in a moving mask (of
size 3 x 3 in our case) resulting in more homogeneous objects. (We preferred median
filtering for its relative simplicity and effectiveness in case of multiplicative noise.)
2. Image Segmentation. Once the image has been smoothed, it is subjected to a segmentation process that separates floes from the rest of the image. This segmentation
technique is based on local dynamic thresholding (Haverkamp et al. 1995), which has
been proven to be able to detect local details of the segmentation and be adaptive to
different SAR sea ice imagery where intensities of similar ice classes vary. Figure 6 shows
Fig.6. Histograms of an ERS-l
SAR sea ice image. Black: The
original image intensity. Red:
The intensities of floes. Green:
The intensities of water/ice
mixture regions. Blue: The
intensities of water regions. Yellow, cyan, and white: Intersections of two or three different
regions. The figure clearly
shows that intensities of different regions overlap.
1 We will use "substrate" and "water/ice mixture" interchangeably for labeling purposes in our figures.
L.-K. SOH, C. TSATSOULIS, AND B. HOLT
tering, (5) separation of open water from the remainder of the ice cover not identified
as floes, termed water/ice mixture extraction, and (6) computation of floe size distribution and coverage statistics. Discussing stage 5) further, after floe extraction and filtering, much of the ice cover is not identified as distinct floes. This primarily includes
pieces of ice that are on the order of the SAR resolution (variable but generally 100 m
or less) and new ice forms. We have termed this remaining ice water/ice mixture' for
convenience. With the knowledge of the floe, water/ice mixture, and water distributions,
we are able to compute complete coverage statistics within a SAR sea ice image. Section 2.5 below discusses an application for examining summer ice where knowledge of
the quantities of identifiable floes and their sizes, open water, and the remaining
water/ice mixture are important.
1. Image Enhancement. To ensure well-defined floes and reduction of radar speckle noise effects (Ulaby et al. 1982), we use median filtering (Huang et al. 1979). Median
filtering enhances the image by eliminating maxima or minima in a moving mask (of
size 3 x 3 in our case) resulting in more homogeneous objects. (We preferred median
filtering for its relative simplicity and effectiveness in case of multiplicative noise.)
2. Image Segmentation. Once the image has been smoothed, it is subjected to a segmentation process that separates floes from the rest of the image. This segmentation
technique is based on local dynamic thresholding (Haverkamp et al. 1995), which has
been proven to be able to detect local details of the segmentation and be adaptive to
different SAR sea ice imagery where intensities of similar ice classes vary. Figure 6 shows
Fig.6. Histograms of an ERS-l
SAR sea ice image. Black: The
original image intensity. Red:
The intensities of floes. Green:
The intensities of water/ice
mixture regions. Blue: The
intensities of water regions. Yellow, cyan, and white: Intersections of two or three different
regions. The figure clearly
shows that intensities of different regions overlap.
1 We will use "substrate" and "water/ice mixture" interchangeably for labeling purposes in our figures.