An input dependent algorithm for the inverse discrete wavelet transform

We propose a fast algorithm for computing the inverse discrete wavelet transform (IDWT). The method takes advantage of the large number of zero coefficients after quantization. A bit map of the wavelet coefficients is used to test for streams of zero coefficients and inverse transform filtering is omitted for these coefficients. The zero testing algorithm is a tree-like search operation. Information on the statistics of the wavelet transform coefficients of a few typical images, together with estimates of the cost of testing and the cost of filtering, are used to obtain optimal sizes of the root and leaves of the search tree. Results show that our algorithm is faster than the baseline inverse wavelet transform algorithm by about 20% to 50% for PSNRs in the range 35 dB to 29 dB.