A talk by Andrew Jupe as part of the Computing Department’s Work in Progress seminar series.
An obvious problem in texture classification is that even a previously-known texture may be difficult to recognise if it does not have the expected orientation. A published method for the re-orientation of image sub-windows will be described; the authors hint that a modification to their method may be worthwhile, and so a modification has been devised. The angle through which the sub-window is rotated is selected via examination of the Radon transform. Following this, recursive wavelet decomposition is applied in order to extract a 30-element feature vector from the image, and then k-nearest neighbour classification is applied.