Image model and spectral extrapolation in transform image coding
Hemant B. Kekre, Jitesh Solanki · International Journal of Electronics · 1978
The philosophy of transform image coding is to transmit only relatively large magnitude transform coefficients and discard the remainder, The picture received is distorted, due to the presence of various noise sources at different stages of transmission. It is possible to design a non-recursive Wiener filter for the information transmission system from a knowledge of statistics of the image and the noise sources, as well as the interaction between them. In defining models, one should incorporate into them as much a priori knowledge about the system as possible. The success of the design is intimately tied to the accuracy of the models. In this paper the problem of spectral extrapolation, in which the rejected transform coefficients are estimated using statistical and correlation techniques, is solved by using different image models. The extrapolation results of the isotropic image model are much superior compared to the separable first-order Markov process model.