UBC Theses and Dissertations
Three-dimensional transform encoding and block quantization of still colour and monochrome moving pictures Soubigou, André
A bandwidth compression scheme for digital colour images using three-dimensional Fourier transform encoding and block quantization is investigated. The brightness U(x,y,λ) of a colour picture is considered as a three-dimensional function of the spatial dimensions x,y and the wavelength dimension λ. A set of colour pictures has been sampled. The study of the second order statistics of the data indicates various models for the autocorrelation function, depending on the type of picture. A separable exponential model is shown to provide a reasonable fit to strongly correlated data having vertical and horizontal features. When those features are not present a non-separable exponential model provides a better fit. Many experimental results of the three-dimensional Fourier transform encoding and block quantization of colour pictures are presented. The effects of the sub-picture bize and the average number of bits per picture element (b.p.p.e.) are considered; 2.75 b.p.p.e. seems sufficient to reproduce a reasonably good quality picture. The use of the separable model is shown to yield results close to the optimum possible at low bit rates, although the model fails to closely predict the signal-to-noise ratio of the reconstructed picture. It is shown that a non-separable model yields better prediction, at -considerable computational expense, but it does not significantly improve the actual performance. An adaptive system is also considered. The effect of digital channel errors on three-dimensional transform encoded pictures is investigated. The application of three-dimensional transform encoding for monochrome time-varying pictures is also considered. Fourier and Hadamard transformations are shown to yield similar results for the specific set of pictures used in our study. Included in the thesis is a summary of results and suggestions for further work.
Item Citations and Data