Linear Models in Spatial Discriminant Analysis
This paper reports on developments of a linear model, proposed in Haslett and Horgan (1985), for the reconstruction of binary (2 colour) images corrupted by additive Gaussian noise. The proposed method involved constructing a simple linear filter of the image, and thresholding the result. This very simple method was competitive with others, much more complicated, in terms of accuracy and speed. Further, by virtue of its close affinities with classical statistical methods of multivariate linear discriminant analysis, it yielded important and easily interpretable qualifications to any given reconstruction, notably (a) an estimated value for the proportion of pixels correctly classified and (b) posterior probabilities for the colour of each pixel.
KeywordsLinear Discriminant Analysis Discriminant Function Neighbourhood Size Noisy Image Additive Gaussian Noise
Unable to display preview. Download preview PDF.
- Besag, J., “On the Statistical Analysis of Dirty Pictures”, read to the Royal Statistical Society, May 1986, and to appear in JRSS(B).Google Scholar
- Haslett, J. and G. Horgan, “Spatial Discriminant Analysis — A Linear Function for the Black/White Case”, presented to SERC Workshop “Statistics and Pattern Recognition”, Edinburgh 1985, and submitted for publication.Google Scholar
- Kent, J.T. and K.V. Mardia, “Spatial Classification Using Fuzzy Membership Models”, preprint, 1986.Google Scholar
- Mardia, K.V., J.T. Kent and J.M. Bibby, “Multivariate Analysis”, Academic Press,Google Scholar
- Saebo, H.V., K. Braten, N.L. Hjort, B. Llewellyn and E. Mohn, “Contextual Classification of Remotely Sensed Data: Statistical Methods and Development of a System”, Rept. 768, Norwegian Computing Centre, 1985.Google Scholar