Two-dimensional dyadic subband tree modeling with embedded quantizers

We represent the effects of scalar quantization by a nonlinear gain-plus-additive-noise model for the PDF-optimized quantizer in a 2-D dyadic subband tree structure. Based on the equivalent models and the polyphase decomposition approach, we compute the complete mean square error (MSE) using cyclostationary concepts. Then, the optimal filter coefficients satisfying paraunitary or biorthogonal conditions, bit allocations and compensation vectors are obtained such that the above MSE is minimized. This design procedure is then used in compressing the Lena image.