Design of a tree-structured bank with allpass polyphase filters for subband image coding via time-frequency localization

We discuss the optimization of filter coefficients in a tree-structured bank with allpass polyphase filters for subband image coding. First we show that the impulse response length and cutoff characteristics of the filter bank have a tradeoff relationship in terms of coding performance. Then we propose a method of designing a bank by minimizing the product △n△w between the time and bandwidths of subband filters. Amplitude responses of the designed subband filters nearly approximate Gaussian functions and their △n△w are little larger than the lower bound of the uncertainty principle. Coding simulations show that the coding performance of the designed bank compares favorably with that of image-optimized banks.