Optimum frequency band partition for subband image coding

Iteration formulas for determining the optimum frequency band partitions for subband image coding are derived theoretically. Numerical examples of the optimum band partition are shown for the first-order Markov signal with a correlation coefficient of near unity, since image signals have nearly equal correlation characteristics to this signal on the average. In addition, data compression performances for the optimum band partition are compared with those for the linear and the log-linear band partition. From the numerical results, it is shown that the optimum partition for the image is near the partition by the wavelet transform.<>