February Fourier Talks 2008

Wavelet filter transforms in detail

One of the more efficient implementations of a perfect reconstruction wavelet filter transform is factorization into Sweldens' lifting steps, which incidentally performs the evaluation in place and reduces memory use. A recent result of Wei ZHU shows that in general these steps need only use nearest-neighbor data. We describe the proof and discuss the advantages and disadvantages from the implementation and accuracy points of view.