Liverpool John Moores University
|
Liverpool John Moores University
|
|
Image Processing Automated analysis of acoustic images from boreholes *K. Glossop, P.J.G. Lisboa, P.C. Russell** and Professor G.R. Jones** *Robertson Geologging **Dept. Elec. Eng., University of Liverpool Geophysical analysis of reflected ultra-sound images from boreholes is an important prospective application with a strong world-wide market. It relies currently on heavy operator interaction, which is expensive and time consuming. The project combines novel developments in image processing methods together with the use of intelligent post-processing systems, to automatically detect and label a wide range of rock fractures directly from the images collected, leaving experienced operators to deal only with the more complex geological features. Teaching Company Scheme with Robertson Geologging in Deganwy, North Wales.
Sample of a borehole image trace following coding to identify strata types and directional information.
Sinusoidal Hough-transform developed for the automatic identification and coding of fracture boundaries.
Phase unwrapping in fringe analysis using wavelet filters S. Hamzah*, C.A. Hobson*, J.D. Pearson* and P.J.G. Lisboa *School of Electrical Engineering, Electronics and Physics, Liverpool John Moores University Wavelet filters are applied to single 2-D projections of 3-D images, acquired under structured light, in order to successfully recover their 3-D profile.
Illustration of the use of structured light to detect 3-D structure from a single 2-D image.
Schematic of the phase unwrapping process, to which wavelet filtering is applied.
Face recognition modelled on bottom-up attentional mechanisms in early vision M.A. Grudin*, D.M. Harvey* and P.J.G. Lisboa *School of Electrical Engineering, Electronics and Physics, Liverpool John Moores University The bandpass filtering carried out by simple cells in the primary visual cortex is simulated in software, using pyramids of Gabor wavelets. Each filter bank is attached to a node in a Dynamic Link Map, which maps preferentially onto new images of the corresponding subject. This image representation has considerable robustness against elastic distortions, 2- and 3-D rotations, and scaling transformations.
Dynamic Link Map applied to an original and distorted image of a subject used to train the database.
New maps are created from a single image of subjects to be introduced into the database, shown on the left. These maps adjust to new presentations, correctly identifying the subject, which is illustrated on the right. |
