While doing my PhD at the Oceans Systems Lab at Heriot Watt University in Edinburgh, Scotland, I looked into developing a system able to detect the presence of man-made objects in underwater images (or videos).
First, the underwater images are restored. Second, the system determined automatically an optimal scale for contour extraction by optimising a quality metric. Third, a Bayesian classifier determined whether the image contained man-made objects (e.g. pipelines). The features used by the classifier were based on the regular shape properties of man-made objects (using measures inspired by perceptual organisation). The system correctly classified approximately 85% of 1390 test images from five different underwater videos in spite of the varying image contents and poor quality. The images below present an example of the results obtained with the underwater image restoration algorithm.
Trucco E. and Olmos, A. (2005), "Self-tuning underwater image restoration ", IEEE Journal of Oceanic Engineering, vol. 31, no. 2, pp. 511-519, 2006.
Olmos, A.Trucco, E. and Lane, D. (2002), " Automatic man-made object detection with intensity cameras (using Support Vector Machines) ", Oceans 2002 MTS/IEEE Conference, p. (1555 - 1561) Vol. 3, October 2002.
Olmos, A. and Trucco, E. (2002), " Detecting man-made objects in unconstrained subsea videos ", Proceedings of the British Machine Vision Conference, pp. 517-526, 2002.
Olmos, A. Trucco, E., Lebart, K., Lane, D., (2000), " Detecting ripple patterns in mission videos ", MTS/IEEE OCEANS 2000, p. (331-335) Vol. 1, September 2000.