An Extension of 3D Zernike Moments for ShapeDescription and Retrieval of Maps Defined inRectangular Solids

Atilla Sit, Julie C. Mitchell, G.n. Phillips, Stephen J. WrightComputational 2013 · DOI

Introduction

3D Zernike moments have long served as a cornerstone for shape description and retrieval in fields ranging from computer graphics to structural biology. These moments rely on a set of orthonormal polynomials that provide a compact and efficient way to represent 3D objects. However, a fundamental limitation of classical Zernike polynomials is their definition over the unit ball. While this is ideal for spherical data, it creates significant challenges for data naturally stored in rectangular volumes, such as density maps from the Electron Microscopy Data Bank (EMDB).

The Challenge: The "Corner Problem"

When standard Zernike polynomials (referred to here as "BallZern") are applied to data within a rectangular prism or cube, their orthonormality property—the mathematical independence of each expansion term—breaks down. This leads to the "corner problem": the spherical basis functions are effectively blind to the information located in the corners and along the edges of the rectangular domain. To capture these details, an impractically high number of expansion terms is required, making the representation inefficient for high-resolution 3D solids.

Methodology: Re-orthogonalizing the Basis

To address this, the researchers developed a method to extend Zernike moments to rectangular domains by re-orthogonalizing the polynomial basis over a specified box-shaped region $\mathcal{R}$. This is achieved through a Gram-Schmidt process, transforming the classical polynomials into a new basis, termed "CubeZern," which is specifically optimized for rectangular geometry.

The paper highlights two implementation strategies:

  1. Recursive Gram-Schmidt: A traditional step-by-step construction.
  2. Non-recursive Matrix Approach: A more numerically stable method utilizing Cholesky factorization of the inner product matrix. This approach is critical for high-order expansions, as it prevents the numerical divergence and loss of orthogonality common in recursive implementations.

Experimental Results and Performance

The CubeZern basis was rigorously tested against standard BallZern moments and simple voxel-based compression using 3D density maps from the EMDB, including primary test cases like EMD-5167 (a microtubule-kinesin complex).

Comparison of 3D reconstructions for EMD-5167 using BallZern and CubeZern bases at expansion order N=20.

Quantitative Improvements

Across various grid sizes ($48^3$ to $128^3$), CubeZern consistently outperformed the classical approach. On the EMD-5167 map, CubeZern achieved a 13–21% lower Root Mean Square Deviation (RMSD) and a 5–7% higher correlation (CORR) than BallZern. Qualitatively, the CubeZern reconstructions were significantly more effective at retrieving density near the volume's boundaries, where BallZern typically fails.

Quantitative evaluation showing RMSD and Correlation improvements across different expansion orders for the CubeZern basis.

Efficiency vs. Compression

When compared to standard map compression (downsampling), CubeZern proved to be a more information-dense descriptor. For an equivalent number of coefficients, CubeZern maintained a 2.5% higher correlation and a 12.5% relative decrease in RMSD, demonstrating its utility as a high-fidelity "fingerprint" for 3D shapes.

Discussion and Future Outlook

While CubeZern offers superior accuracy for rectangular data, it introduces a trade-off: the loss of the rotational invariance inherent in standard Zernike descriptors. Because the basis is tied to the box geometry, the resulting moments are sensitive to the object's orientation.

Future research aims to bridge this gap by deriving rotationally invariant descriptors from the CubeZern basis. Additionally, optimizing computational workflows for very high-order expansions will be necessary to handle the increasing resolution of modern biological imaging datasets. Despite these hurdles, the extension of Zernike moments to rectangular domains represents a significant step forward in the efficient representation of complex 3D solids.

Summary updated

Abstract

Abstract Zernike polynomials have been widely used in the description and shape retrieval of 3D objects. These orthonormal polynomials allow for efficient description and reconstruction of objects that can be scaled to fit within the unit ball. However, maps defined within box-shaped regions ¶ for example, rectangular prisms or cubes ¶ are not well suited to representation by Zernike polynomials, because these functions are not orthogonal over such regions. In particular, the representations require many expansion terms to describe object features along the edges and corners of the region. We overcome this problem by applying a Gram-Schmidt process to re-orthogonalize the Zernike polynomials so that they recover the orthonormality property over a specified box-shaped domain. We compare the shape retrieval performance of these new polynomial bases to that of the classical Zernike unit-ball polynomials.
BibTeX
@inproceedings{sit2013extension,
  author = {Atilla Sit and Julie C. Mitchell and G.N. Phillips and Stephen J. Wright},
  title = {{An Extension of 3D Zernike Moments for ShapeDescription and Retrieval of Maps Defined inRectangular Solids}},
  booktitle = {Computational},
  year = {2013},
  doi = {10.2478/mlbmb-2013-0004},
}
Figure