126
6.4
Discussion
A. K. Lm AND C. Y. PENG
In this paper, the two-dimensional Gaussian wavelet transform of SAR images has been
used with transforms of various scales to separate texture or features. The use of
wavelet transforms for tracking ice features in the SAR imagery has been demonstrated, especially in situations where the feature correlation techniques such as the ones
used in the ASF Geophysical Processor System fail to yield reasonable results. The evolution of the St. Lawrence polynya in the Bering Sea was studied using ERS-l SAR data
with a 3-day repeat cycle. This wavelet analysis procedure is able to systematically
extract geophysically relevant parameters from SAR imagery of polynyas. The procedure is robust and can contribute to the monitoring of polynyas with quantitative information on a large scale.
For the St. Lawrence polynya study, the evolution of the polynya can be easily tracked
by wavelet analysis due to the large difference in backscatter between the water and the
ice. The "proximity to an approximation" method has been developed in order to locate
a more accurate boundary through the use of a distance criterion. With the ice edge
detected from the large-scale wavelet transform used as the approximate boundary, the
edge elements from the small wavelet transform are then used as boundary points for
a more accurate boundary. Several scales can be used to delineate the ice edge as accurately as required. Therefore, wavelet analysis of SAR images may be useful for largescale polynya monitoring. The procedure of ice floe tracking near the polynya is similar to tracking of the ice edge. In the case of St. Lawrence Island, first the wavelet transform of SAR images is computed with a median scale, corresponding to typical ice floe
size in this area. Then the method of XOR is applied for template matching and the
matched ice floes are used for tracking. The detailed shape of an ice floe can be constructed by assembling edge elements obtained by wavelet transform using a small scale.
Open leads (dark areas) can be tracked similarly by using the two-dimensional
wavelet transform with a small scale as an edge detector to separate the leads areas and
produce a binary image as shown in Fig. 2. In this case a threshold below zero will be
used to delineate the dark leads area. In addition, a small scale of 2 units of pixel spacing will be used for sharp boundaries between water and ice. Directional wavelet transform, such as the Morlet wavelet, may be very useful for leads aligned in a certain direction. Further study is underway and will be reported separately in the future.
In this study, wavelet analysis was applied primarily for feature or pattern recognition,
and therefore it was fairly insensitive to the absolute calibration of the SAR data. Note that
the present method of template matching of ice floe shapes using a window of a binary
image obtained after manipulating the results of the wavelet transform is very efficient
computationally. This is mainly due to the fact that the only computation operations
involved are bitwise logical operations and additions, while classical template matching
involves the operations of addition, squaring, and multiplication. Furthermore, it is only
necessary to match the template pattern to a limited number of target patterns generated by the results of the wavelet transform, not to every location in the images as with classical template matching. It should also be noted that, although template correlation was
applied here only to fmd the translation of the target pattern with respect to the template
pattern, it can be extended to find the rotation of the target pattern by incremental rotation of the target pattern and then matching the degree of agreement.
6.4
Discussion
A. K. Lm AND C. Y. PENG
In this paper, the two-dimensional Gaussian wavelet transform of SAR images has been
used with transforms of various scales to separate texture or features. The use of
wavelet transforms for tracking ice features in the SAR imagery has been demonstrated, especially in situations where the feature correlation techniques such as the ones
used in the ASF Geophysical Processor System fail to yield reasonable results. The evolution of the St. Lawrence polynya in the Bering Sea was studied using ERS-l SAR data
with a 3-day repeat cycle. This wavelet analysis procedure is able to systematically
extract geophysically relevant parameters from SAR imagery of polynyas. The procedure is robust and can contribute to the monitoring of polynyas with quantitative information on a large scale.
For the St. Lawrence polynya study, the evolution of the polynya can be easily tracked
by wavelet analysis due to the large difference in backscatter between the water and the
ice. The "proximity to an approximation" method has been developed in order to locate
a more accurate boundary through the use of a distance criterion. With the ice edge
detected from the large-scale wavelet transform used as the approximate boundary, the
edge elements from the small wavelet transform are then used as boundary points for
a more accurate boundary. Several scales can be used to delineate the ice edge as accurately as required. Therefore, wavelet analysis of SAR images may be useful for largescale polynya monitoring. The procedure of ice floe tracking near the polynya is similar to tracking of the ice edge. In the case of St. Lawrence Island, first the wavelet transform of SAR images is computed with a median scale, corresponding to typical ice floe
size in this area. Then the method of XOR is applied for template matching and the
matched ice floes are used for tracking. The detailed shape of an ice floe can be constructed by assembling edge elements obtained by wavelet transform using a small scale.
Open leads (dark areas) can be tracked similarly by using the two-dimensional
wavelet transform with a small scale as an edge detector to separate the leads areas and
produce a binary image as shown in Fig. 2. In this case a threshold below zero will be
used to delineate the dark leads area. In addition, a small scale of 2 units of pixel spacing will be used for sharp boundaries between water and ice. Directional wavelet transform, such as the Morlet wavelet, may be very useful for leads aligned in a certain direction. Further study is underway and will be reported separately in the future.
In this study, wavelet analysis was applied primarily for feature or pattern recognition,
and therefore it was fairly insensitive to the absolute calibration of the SAR data. Note that
the present method of template matching of ice floe shapes using a window of a binary
image obtained after manipulating the results of the wavelet transform is very efficient
computationally. This is mainly due to the fact that the only computation operations
involved are bitwise logical operations and additions, while classical template matching
involves the operations of addition, squaring, and multiplication. Furthermore, it is only
necessary to match the template pattern to a limited number of target patterns generated by the results of the wavelet transform, not to every location in the images as with classical template matching. It should also be noted that, although template correlation was
applied here only to fmd the translation of the target pattern with respect to the template
pattern, it can be extended to find the rotation of the target pattern by incremental rotation of the target pattern and then matching the degree of agreement.