Addendum to: A new median formula with applications to PDE based denoising

Yingying Li, Stanley OsherCommunications 2010 · DOI

Introduction

This document serves as a formal addendum to the research paper "A New Median Formula with Applications to PDE Based Denoising," originally published in Communications in Mathematical Sciences (2009) by Yingying Li and Stanley Osher. The purpose of this addendum is to acknowledge concurrent, unpublished research by Adam Oberman and to clarify the mathematical positioning and scope of the authors' original contributions within the field of image processing.

The Landscape of PDE-Based Denoising

Image denoising remains a fundamental challenge in computer vision, primarily centered on the removal of unwanted artifacts while preserving critical structural features like edges. Total Variation (TV) denoising has long been a standard approach to this problem; however, TV methods often require complex optimization techniques.

The work of Li and Osher explores "median formulas," which act as a mathematical bridge between simple local filters—such as the median filter, known for its effectiveness against "salt-and-pepper" noise—and the more sophisticated variational frameworks of Partial Differential Equations (PDEs). By providing a rigorous formula that describes how these median-based operations evolve an image over time, the authors offer a path toward more efficient and mathematically grounded denoising algorithms.

Methodological Framework

The core methodology involves a variational framework focused on the minimization of specific functionals. The authors specifically analyze cases where the functional takes the form $F(x) = (x-u)^2$. Their primary contribution is a concise formula representing the solution to these PDEs as a local median operation.

A significant aspect of this research involves the transition from discrete to continuous models. In digital imaging, denoising schemes are often implemented using anisotropic discretized forms, where a pixel's update is determined by its interaction with its immediate neighbors (typically a 4-neighbor or 8-neighbor grid). The authors demonstrate that these common discrete schemes are manifestations of the broader continuous PDE theories they have developed.

Generalization and Scholarly Context

The primary motivation for this addendum is to address the "scholarly landscape" regarding polyhedral total variation and specific discretized schemes. Through private communication, the authors became aware of unpublished work by Adam Oberman (2006) that proposed similar schemes for TV denoising using polyhedral norms.

Li and Osher clarify that while Oberman independently discovered specific cases of these formulas—particularly those involving the squared difference functional—their own work provides a more generalized and concise mathematical framework. By encompassing Oberman’s results as special cases, the authors reinforce the robustness of their theorem, which extends to a broader class of functional specifications than previously established discrete schemes. In theoretical mathematics, such conciseness is a virtue, as it often reveals the underlying symmetry or fundamental nature of a problem more clearly than case-by-case derivations.

Conclusion

This addendum reinforces the significance of the original 2009 paper by situating it as a generalized evolution of existing discrete denoising methods. It acknowledges the important concurrent work of Adam Oberman while asserting the broader applicability and mathematical elegance of the authors' unified median formula. The work remains a cornerstone for understanding the link between local discrete filters and global variational PDE models.

Summary updated

BibTeX
@inproceedings{li2010addendum,
  author = {Yingying Li and Yingying Li and Stanley Osher and Stanley Osher},
  title = {{Addendum to: A new median formula with applications to PDE based denoising}},
  booktitle = {Communications},
  year = {2010},
  doi = {10.4310/cms.2010.v8.n4.a16},
}

Loading related content...

Figure