Abstract: We derive the relationship between the persistent homology barcodes of two dual filtered CW complexes. Applied to greyscale digital images, we obtain an algorithm to convert barcodes between the two different (dual) topological models of pixel connectivity.
TL;DR
This paper establishes a formal mathematical bridge between the two primary ways of representing digital images for topological analysis. By proving a duality theorem for persistent homology, the authors provide an algorithm that allows researchers to compute topological features (barcodes) for one image model and instantly "translate" them to the other, significantly reducing computational overhead.
What Problem Does This Paper Address?
In the field of Topological Data Analysis (TDA), researchers use persistent homology to find "holes" in data—such as connected components, loops, or voids—that persist across different intensity thresholds. When applying this to greyscale images, there are two standard ways to build a mathematical model (a "complex"):
- The T-construction: Pixels are treated as solid squares (or cubes in 3D).
- The V-construction: Pixels are treated as vertices (points) connected by edges.
These two models represent different "connectivity" rules (direct vs. indirect adjacency). Historically, if a researcher wanted to understand an image through both lenses, they had to run two separate, computationally expensive algorithms. Furthermore, while a relationship between these models was suspected to be a form of Alexander Duality, a rigorous proof and a general conversion algorithm for cell-based complexes were missing.
Methods
The authors leverage the framework of CW complexes (a general way to build spaces by gluing "cells" of different dimensions) and Discrete Morse Theory to simplify the underlying mathematics.
The core of their approach involves:
- Defining Dual Filtered Complexes: They define what it means for two filtered complexes to be dual, requiring that their cells correspond, their adjacency relations are reversed, and their filtration orders (the sequence in which pixels "appear") are flipped.
- One-Point Compactification: To handle the edges of an image (where duality usually breaks down), they "pad" the image with a layer of pixels at infinity and mathematically glue the boundary to a single point, turning the flat image into a topological sphere.
- Discrete Morse Theory: They use this to reduce the complexity of the chain complexes, making it easier to track how "critical cells" map between the dual representations.
Datasets
As a theoretical and algorithmic paper, it does not utilize large-scale external benchmarks. Instead, it uses synthetic greyscale arrays. Small $3 \times 3$ and $2 \times 2$ pixel arrays are used to provide step-by-step visualizations of how the T- and V-constructions differ and how the duality theorem maps their barcodes.
Key Insights and Ideas
- The "Aha" Moment on Dimensionality: One of the most powerful implications of this duality is that a high-dimensional feature in one construction corresponds to a low-dimensional feature in the other. For example, finding 2D "voids" in a 3D image can be achieved by finding 0D "connected components" in the dual image, which is computationally much faster.
- Sign-Flipped Persistence: The paper shows that the "birth" and "death" times of a topological feature in one model are simply the negative "death" and "birth" times in the dual model.
- Connectivity is Symmetric: The paper proves that the "direct adjacency" (open topology) of the V-construction and the "indirect adjacency" (closed topology) of the T-construction are mathematically inseparable—they are two sides of the same coin.
Key Results
The paper culminates in two major theorems:
- Theorem 2.1: Provides a general bijection for the persistent homology barcodes of any two dual filtered CW complexes. It shows that a feature $[p, q)$ in dimension $k$ maps to a feature $[-q, -p)$ in dimension $d-k-1$ of the dual.
- Theorem 3.1: Tailors this result specifically to digital images. It provides a concrete "recipe" for converting barcodes between the T-construction and V-construction.
This means software designed for one construction can now be used to get results for the other without any new topological code—just a simple post-processing step on the output barcodes.
Limitations and Future Work
- Extended Abstract: This paper is a summary of ongoing research; while the core proofs are sketched, the full technical depth of the chain isomorphisms is abbreviated.
- Manifold Restrictions: While the results generalize previous work that was limited to spheres, the current application focuses heavily on cubical complexes (images).
- Future Directions: The authors suggest that this duality can be used to adapt advanced image processing techniques, like skeletonization and partitioning, to work more efficiently by switching to the dual representation when it is computationally cheaper.
Summary updated