Construction of symmetric fractional over-complete wavelets and applications in image restoration

摘要

In this work, a novel design scheme is proposed for the construction of symmetric fractional over-complete wavelet filter banks. We first provide solutions to the open problem of designing low-pass filters that are symmetric and of minimum-length. We then obtain the high high-pass filters via Toeplitz matrix factorization which is of less computational complexity than existing methods. The resulting filter banks are approximately shift-invariant. The designed filter banks are applied in image restoration that uses an analysis based model solved by split Bregman algorithms. The experiments show the constructed symmetric fractional over-complete wavelet transforms (FOWTs) allow better restoration results than some other wavelet transforms in the literature.