|
Digital Image Processing
|
Automatic Detection of V-Shaped Features from SAR Imagery
4. Results and Discussion
Figure 5 shows the final extracted V-shaped ship wakes. The Information Fusion Apprch worked farily well to detect V-shaped thin line features which are only manor importance in the image. A very clear upward pointing ship wake in the lower left of the image and a V-shape wake for the ship in the top right corner of the image were successfully extracted. No ship wakes could be found for the ship in the middle, as it had no ship wake visible in the original image. As well, the wakes for the downward ship in the bottom left of the image were not extracted probably because it does not form a clear V-shape.
Figure 5 Ship wakes extracted from the ERS-1 image
The algorithm was tested on a 1000 line 4 - look RADARSAT image of the Atlantic Ocean southwest of Wales, UK taken on February 29, 1996 shown in Figure 6. the image shows two ships and the corresponding wakes visible to the naked eye, both traveling in a generally upward direction. The background water is fairly rough and the ship wakes are not very obvious features. The result for the RADARSAT image shown in Figure 7 is very good without a false alarm. However, the wakes extend to the other direction because of some small bright features on the line. It clearly shows that at least one more evaluation step is necessary after extracting intersecting line segments to remove those extra features. It can done by evaluating the pixel intensity distribution of the all branches of intersecting line segments or/and by evaluating the intersecting angle.
Figure 6 RADARSAT image of Atlantic ocean southwest of Wales, UK.
Figure 7 Ship wakes extracted from the RADARSAT image
The results may be further improved by the following consideration.
-
Integration of more knowledge. The search space of the ERS-1 image is further reduced by integrating the ship route. This will certainly eliminate most of the confusing line-like features on the rough water area away from the shore. Those must be some water area for feisue boats. The similar improvement can be expected for the RADARSAT image if such a map is available.
- Localized Hough transform space. In this experiment, Hough transform was applied to the entire image plane. If Hugh transform was applied in the subspace around candidate line-like features, it will enhance the accuracy but also reduce the number of confusing feature.
- Additional simple spatial operation to simulate the spatial elongatedness. Only three steps were applied in the pre-processing to eliminate unwanted features, additional processes such as estimating a medium range elongated spatial distribution. These operations could further reduce the number of confusing features.
On of the most interesting areas of investigation was the use of Genetic Algorithms to find peaks in Hough space. The Genetic Allgorithms peak finding was run on two of the Hough images: the averaged image and difference image. the results from difference image was somewhat disappointing as the righ-hand wake of the upward bottom left ship wake failed to be detected in the top 20 peaks. While the results did not prove to be quite as good as those from the initial peakfinding algorithm, they did provide a very effective option. More investigation, however, is necessary to find the most appropriate input image and fitness function.
5. Conclusion
the "Information Fusion Approach" works quite well to detect insignificant V-shaped features from relatively large image. the system can be further improved by integrating more knowledge and more efficient process. Genetic Algorithms have potential, but do not work as quickly as the image processing approach.
References
-
J.Lisaka and T. Sakurai-Amang, "Circular pattern extraction from remotely sensed imagery using Genetic Algorithms", Proc. GIS'95, pp.382-390, Vancouver, Canada, 1995.
- J.lisaka and T.Sakurai-Amano, "Automated detection of road intersections from ERS-SAR imagery", Proc. IGARSS'95, pp.676-678, Florence, Italy, 1995.
- J.lisaka, T.Sakurai-Amano, S.Moresoli, "Automated road and trail detection from tropical region SAR imagery", PROC. GIS'96, Vancouver, Canada, 1996.
- J.lisaka and T.Sakurai-Amano, "Automated GCP detectionfor SAR imagery. I, Road intersection", Proc. SPIE Conference 2818, Denver, Coloarado, USA, 1996.
- J.lisaka and T.Sakurai-Amano, "PC Network of image computing for remote sensing", Proc. Symp. N global Environmental monitoring from space" pp-56-66Tokyo, Japan, 1984.
- L.M. Murphy, "Linear feature detection and enhancement in noisy images via the Radon transform", Pattern Regognition Lett.,vol.4No. 4,pp.279-284,1984.
- K. Eldhuset. "Automated ship and ship wake detection in spaceborne SAR imagesfrom coastal regions", Proc.-IGARESS'88, Edinburg, Scotland, 1988.
- K.Eldhuset, "Principles and performance of an automated ship detection system for SAR images," Proc. IGARSS'88, Edinburg, Scotland, 1988.
- R.F. Gasparovic, D.R. Thomson, and J.R. Apel, "Synthetic aperture radar imaging of ship-generated internal waves", John Hopkins APL Tech. Dig. Vo; no.4, pp.326-331, 1989.
- M.R. Rey, J.K.E. Tunaley, J.T. Folinsbee, P.A. Jahans, J.A. Dixon, and M.R. Vant "Application of Radon transform techniques to wake detection in SEASAT-A SAR images", IEEEE. Trans. Geosci, Remote Sensing, Vol. 28. pp. 553-560.1990.
- A.C. Copeland, G.Ravichandran, M.H. Trivedi, "Localized Randon transform-based detection of ship wakes in SAR image", IEEE Trnas. Geosci, And Remote Sens., vol.33, no.1pp.35-45, 1995.
- J.lisaka and T.Skurai-Amano, "Terrain featue recognition for SAR imagery employing spatial attributes of targets", Proc ISPRS Commission III Symp. Vo; 2357-399-408, Munich, Germany, 1994.
- J.lisaka and T.Sakurai-Amano, "Automated terrain feature detection from remotely sensed images integrating spectral, spatial and geometrical attributes of object", Proc. GIS/LIS'95, PP. 486-495, Nashville, USA, 1995.
- T.Sakuri-Amano and J.lisaka, "Small feature preserving speckle reduction High level image product", MPO Technical Report, No. 297-94, Canada Centre for Remote Sensing, Ottawa, Canada, 1994.
- J.lisaka and T.Sakurai-Amano, "A shape-based object recognition for remote sensing", Proc.. IGARSS'95, pp. 160-162, Florence, Italy, 1995.
- J.lisaka, "Structural spatial information extraction from remotely sensed data", Proc. IGARSS'89, pp. 1224-1227, Vancouver, Canada, 1989.
- R.O. Duda and P.E.Hart, "Use of the Hough transform applied to detect lines and curves in pictures", Comm. ACM, vol. 15, no. 1 pp11-15, 1972.
- D.E. Goldberg, "Genetic Algorthms in Search, Optimization, and Machine Learning, Addison Wesley, New York, 1989.
|
|
|
|
|
