Automated Extraction of Shorelines from Optical and SAR Image
Shoreline Detection from SAR Images
The flowchart of detecting shorelines from SAR images are shown in fig 2. The outlines are described follows.
Fig. 2. The Flowchart of Extracting shorelines from SAR Images
We use Hybrid Sigma filter [Alparone et. al. 1995] to remove the speckles. The filter modified the Lee sigma filter [Lee, 1983] by fluctuating the mean value of Gaussian distribution according to the local window. We then calculate the edge gradients to derive edge strength. Then a suppression for local non-maximum is performed followed by line tracing. At this stage, precision edges are derived. However, not all of them are shorelines. Thus, we need the next procedure.
We test ratio edge detector to segment the blocks. The method was proposed by Touzi et al [1988]. In which, the detector compared the edge ratio probability for a pixel and its neighbors to locate the boundary of a block. Then a morphological thinning [Pratt, 1991] and line tracing are applied to complete the skeletonization. After a buffer zoning, we determine the potential shorelines areas. Through an intersection between the areas and the precision edges generated in the previous procedure, precision shorelines are extracted.
Experimental Results
Two areas were tested. The site is called Wai-Sun-Ding (WSD). The second one is Guan-Ying (GY). A SPOT multi-spectral image and an ERS-1 SAR images were included in both sites. For WSD case, we compared the shorelines digitized manually and
the one extracted automatically. While, for GY site, we compared the results of GPS measurements with respect to the one from images.
Fig. 3 represents the results of classification from the SPOT image. Fig. 4 is the superimposition of the original image and the extracted precision shorelines. The consistency of the shorelines extracted from the proposed scheme and manual digitization is better than 2% in terms of the area. Fig. 5 illustrates the total edges on the ERS-1 image. Fig 6 shows the precision shorelines on the original ERS-1 image. Again, the consistency between the proposed scheme and manual digitization is better than 2%. Fig 7 represents the results of GY case. In which, the measurements. The RMSE is better than 1 pixel. Fig 8 is for evaluation of SAR results. The bright line is the shorelines from ERS-images. The dark segment is from GPS measurement.
The RMSE is 1.5 pixels.
Fig. 3 Results of Classification
Fig. 4 Precision Shorelines on the SPOT Images.
Fig. 5. Total Edges on the SAR Image
Fig. 6. Precision Shorelines on the SAR Image
Fig. 7 SPOT Shorelines vs. GPS Measurements
Fig. 8. SAR Shoreline vs. GPS Measurements
