Skip to Main Content
The wavelet transform, similar to the short-time Fourier transform, furnishes an alternative approach to signal processing, especially suitable for the analysis of spatial and spectral locality. We present an efficient VLSI architecture for 2-D biorthogonal inverse discrete wavelet transforms. Based on the zerotree-like synthesis scheme, the proposed architecture can efficiently process the input signal in real-time. To further minimize our hardware cost, three efficient filter structures are designed. For the computation of an N×N 2-D image with the Daubechies (1992) 9-tap/7-tap filter, this architecture spends near N2 clock cycles, and requires about 7N storage elements, 17 multipliers, as well as 27 adders.