At this point, we can delineate the concave areas in image gray value surface by convoluting image with LoG operation and take the negative portion of resulting image. However, not all of such concave areas are oil slick coverage. In fact, oil film and sea water are electromagnetically different. The gradient of image gray value on the boundary of oil-water is greater than that made by other nature phenomena, for instance, the variation of wind field. Here we used a first order derivertive operator, difference of Gaussian (DoG)[3], to measure the sharpness of the edge on the boundary of the slicks just detected by last procedure. A two dimensional DoG operatior can be expressed as follows:
In this paper, the sharpnessof edge s defined as :
S(x,y) = [GoGx(x,y)* I(x,y)]2 +[DoGy(x,y)* I(x,y)]2
(3)
Where I(x,y) is image gray value on position (x,y) and "*" denotes covolution operator.
To justify whether the detected area is oil slick or not, a criterion should be empiricallychosen. In this paper, we select a thrshold as m+1.75s wher m and s are mean and standard deviation of sharpness over all image pixels, respectively. If the slick boundary pixels whose sharpnes is greater than the selected threshold, they are classified as an oil slick boundary. Finally, if the number of oil slick boundary pixels is greater than half of slick boundary pixels, we say this slick is qualified to be an oil slick. The flow chart of LoG plus DoG for oil slicks detection is shown in Figure 3.
Figure 3: The flow chart of LoG plus DoG for oil slicks detection.
Because the oil slicks vary in size, the window size of operator also has to be adjusted accordingly. In this paper, we apply lower resolution window (larger window size) to detect the oil slicks initially and then apply higher resolution window (smaller window size) to detect finer soil slicks based on the oil slicks detected by coaser resolution. This approach effectively reduces the sea clutter and preserves subtle variation of oil slicks.
(3) Coarse to fine multiplayer oil slicks refinement on edge portion
Once the oil slicks is detected in coarset layer, we can now use LoG to further delineate small variation of oil licks on edge portion in the rest of the layers. Note that the window size here use was empirically set to 19 by 19 pixels. The edge portions represent the buffered zone along the edge of oil slicks by a distance or 4 pixels.
(4) Using region growing as a final refinement
Generally, oil slick is not a completely concave surface on image. There may be some small convex surfaces and their curvature can be very small. Figure 4 illustrates this regard. The shadow areas will not be detected with LoG plus DoG procedure alone. We use region growing method[4] along detected oil slicks to find out all of oil slicks. A fullblown procedure is sketched in Figure 5.
Figure 4: Oil slick is not a completely concave surface on image.
Figure 5: the detailed flow chart of proposed scheme.
