Application of image covariance models to transform coding
R.J. Clarke · International Journal of Electronics · 1984
The success of a transform image coding scheme is dependent on a determination of the statistical properties of the data (measured either in the input or the transform domain) in order to ensure efficient allocation of coding capacity (bits) to significant coefficients, and in this regard simple parametric models have often been employed. In this paper the justification for such an approach is examined, using three widely differing test sources, and spectral responses and transform domain coefficient variances are determined both by direct measurement and via simple covariance models. It is shown that, even when the model is successful in accurately representing the image properties in the data domain, significant differences still exist between measured and predicted transform domain properties and, therefore, that such modelling although widely employed, is at best only an approximate guide to the efficient coding of transform coefficients.