Semi-orthogonal wavelets that behave like fractional differentiators

Dimitri Van De Ville, Thierry Blu, Brigitte Forster, Michaël UnserProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2005 · DOI

TL;DR

This paper introduces a new class of "semi-orthogonal differential wavelets" that more accurately mimic fractional-order differential operators. This improved operator-like behavior allows for a more precise characterization of signal singularities, potentially leading to better signal analysis techniques.

What Problem Does This Paper Address?

The power of wavelets in signal analysis often stems from their ability to act like differential operators. This property is crucial for tasks like detecting discontinuities or reconstructing signals from their features. However, standard wavelets only approximate this behavior. The authors aim to bridge this gap by designing wavelets that explicitly behave like fractional-order differential operators, offering finer control over the differentiation process and potentially leading to more robust signal analysis, especially for identifying sharp changes or singularities. They also address the limitations of existing redundant wavelet transforms, which, while improving analysis, suffer from over-completeness and non-unique inverses.

Methods

The core of the paper is the design of "semi-orthogonal differential wavelets." The strategy involves:

  • Fractional B-splines: Using generalized fractional B-splines as the foundation for the scaling functions. These are defined in the Fourier domain and are inherently linked to fractional differential operators.
  • Semi-orthogonality: Ensuring that the wavelet spaces are orthogonal to each other. This is achieved by carefully selecting a filter, denoted as $Q(z^2)$, which balances the wavelet's approximation properties with its differential behavior.
  • Operator Matching: The design explicitly aims to make the wavelet's frequency response closely match that of a fractional differential operator, particularly at low frequencies. This is achieved by choosing specific forms for $Q(z^2)$, leading to wavelets that are better approximations of the desired differential operator.
  • Efficient Implementation: The proposed wavelet transform is implemented using Fast Fourier Transforms (FFTs), making it computationally efficient.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and demonstrates results using synthetic signals (piecewise polynomial).

Key Insights and Ideas

  • Fractional Differentiation is Key: The paper highlights that by allowing fractional orders for derivatives and shifts, we gain a more nuanced tool for signal analysis.
  • Wavelets as Precise Operators: The authors demonstrate that wavelets can be designed to precisely mimic fractional differential operators, not just approximate them. This is a significant step beyond previous work.
  • Singularity Characterization: A key benefit is that signal singularities are characterized by at most two local extrema per subband. This offers a cleaner and more compact way to represent and analyze signal features.
  • Trade-offs in Design: The paper explores different choices for the design filter $Q(z^2)$, showing how they affect the wavelet's approximation of the differential operator and its support.

Key Results

  • Improved Operator Behavior: The proposed differential wavelets exhibit a frequency response that matches fractional differential operators more closely than standard B-spline wavelets, especially for higher fractional orders.
  • Compact Singularity Representation: For a piecewise polynomial signal, the wavelet decomposition using these differential wavelets shows singularities characterized by one or two extrema per subband, even with a high-order transform.
  • Efficient Implementation: The wavelet transform can be efficiently implemented using FFTs.

Limitations and Future Work

  • Theoretical Proofs: Some proofs for the propositions are deferred to a future paper.
  • Infinite Support: For certain choices of fractional orders and shifts, the corresponding filters may have infinite support, which can be a practical limitation. The authors suggest exploring the design freedom of $Q(z^2)$ to address this.
  • Extension to Higher Dimensions: The paper mentions that the concept can be extended to multiple dimensions using polyharmonic B-splines, which will be studied in more detail in future work.
  • Further Refinement: While the basic differential wavelet is presented, the paper also discusses a method for further refining the operator approximation, though it notes the gains become marginal for increasing orders.

Summary updated

Abstract

The approximate behavior of wavelets as differential operators is often considered as one of their most fundamental properties. In this paper, we investigate how we can further improve on the wavelet's behavior as differentiator. In particular, we propose semi-orthogonal differential wavelets. The semi-orthogonality condition ensures that wavelet spaces are mutually orthogonal. The operator, hidden within the wavelet, can be chosen as a generalized differential operator &part;<sup>&#947;</sup><sub>&#964;</sub>, for a &#947;-th order derivative with shift &#964;. Both order of derivation and shift can be chosen fractional. Our design leads us naturally to select the fractional B-splines as scaling functions. By putting the differential wavelet in the perspective of a derivative of a smoothing function, we find that signal singularities are compactly characterized by at most two local extrema of the wavelet coefficients in each subband. This property could be beneficial for signal analysis using wavelet bases. We show that this wavelet transform can be efficiently implemented using FFTs.
BibTeX
@inproceedings{ville2005semi,
  author = {Dimitri Van De Ville and Thierry Blu and Brigitte Forster and Michaël Unser},
  title = {{Semi-orthogonal wavelets that behave like fractional differentiators}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2005},
  doi = {10.1117/12.614791},
}
Figure