An Improved Image Registration Based on Nonsubsampled Contourlet Transform and Zernike Moments

Rui Ding, Ziyan Song, Jin TangInternational Journal of Signal Processing Image Processing and Pattern Recognition 2016 · DOI

TL;DR

This paper presents an improved image registration algorithm that enhances feature extraction and matching for images with significant scaling, rotation, and translation. By refining the use of Nonsubsampled Contourlet Transform (NSCT) for feature point detection and Zernike moments for matching, the authors achieve higher accuracy and robustness, particularly for challenging image pairs.

What Problem Does This Paper Address?

Image registration is the process of aligning two or more images of the same scene. A common and efficient approach is feature-based registration, which relies on identifying and matching distinctive points (features) in images. This method typically involves three steps: feature extraction, feature matching, and transformation parameter estimation. While existing methods using Nonsubsampled Contourlet Transform (NSCT) for feature extraction and Zernike moments for matching have shown promise, they struggle with images exhibiting large transformations (scaling, rotation, translation) and can be prone to errors. Specifically, the original NSCT-based feature extraction can lose rotational invariance, and Zernike moment matching can be unreliable due to sensitivity to the choice of neighborhood size and potential for mismatches. This paper aims to address these limitations by proposing improvements to both the feature extraction and feature matching stages.

Methods

The paper introduces two key improvements to an existing NSCT and Zernike moments-based image registration algorithm:

  • Improved Feature Points Extraction (NSCT):
    • Circular Neighborhood for Local Maxima: Instead of using a square neighborhood to find local maxima of NSCT coefficients, the authors propose using a circular neighborhood. This change significantly improves rotational invariance, ensuring that feature points remain consistent even after image rotation.
  • Boundary Point Elimination: Feature points too close to the image boundary (within a distance 'w') are eliminated. This helps to avoid spurious matches and improve the reliability of extracted features.
  • Improved Feature Points Matching (Zernike Moments):
    • Dual Concentric Circular Neighborhoods: The matching process now utilizes two concentric circular neighborhoods of different sizes. A larger neighborhood is used for initial matching to find candidate pairs, and then a smaller neighborhood is used to refine and verify these matches. This approach enhances the reliability and effectiveness of Zernike moment-based matching.
  • Stability Measurement: The algorithm incorporates a measure of matching point stability, defined as the ratio of the maximum to sub-maximum correlation coefficients. This helps to select more robust matching pairs, especially for images with complex characteristics or large transformations. The top two or three most stable matching pairs are then used for parameter estimation.

Datasets

The paper uses standard grayscale images for experimental evaluation:

  • "Lena" image
  • "cameraman" image
  • "ny_256" image (described as having less obvious characteristics and thus posing greater registration difficulty)

All images are of size 256x256 pixels. The experiments involve creating distorted versions of these images by applying known transformations (scaling, rotation, translation).

Key Insights and Ideas

  • Rotational Invariance is Crucial for Feature Extraction: The paper highlights that a square neighborhood for local maxima detection in NSCT coefficients is insufficient for maintaining rotational invariance, leading to lost correspondences in rotated images. Switching to a circular neighborhood is a simple yet effective solution.
  • Dual Neighborhoods Enhance Matching Robustness: Using a two-stage matching process with different neighborhood sizes for Zernike moments provides a more robust similarity measure, reducing mismatches caused by scale or rotation.
  • Stability as a Key Metric: Incorporating a stability measure for matching points is vital for selecting reliable pairs, especially when dealing with challenging registration scenarios. This goes beyond just correlation and considers the distinctiveness of a match.
  • Improved Performance for Challenging Scenarios: The proposed improvements are particularly beneficial for images with significant scaling, rotation, and translation, where traditional methods often fail.

Key Results

The experimental results demonstrate significant improvements in both the number of correct matching points and the overall matching accuracy:

  • Increased Correct Matching Points: The improved NSCT-based feature extraction method leads to a higher number of correct matching points, especially when rotation is involved. For example, in Figure 7, the "lena" image shows an increase in correct matching points pairs as the rotation angle increases.
  • Higher Matching Accuracy: The improved algorithm consistently achieves higher matching accuracy across various experimental conditions, as shown in Figure 7 and Table 1. For instance, in Table 1, the "Matching accuracy" for the improved algorithm is often 1 or significantly higher than the original method.
  • Robustness to Transformations: The improved algorithm successfully registers images with large scaling (up to 1.4 times), rotation, and translation. Table 3 shows that the estimated transformation parameters are very close to the true parameters, even for significant transformations.
  • Improved Matching Correctness: Table 2 shows that the improved method successfully screens out correct matching points pairs (marked as 'T') where the original method often failed (marked as 'F'), particularly for images with no obvious characteristics or large transformations.

Limitations and Future Work

  • Computational Cost: While not explicitly detailed as a limitation, the use of NSCT and Zernike moments, especially with multiple neighborhood sizes, can be computationally intensive. The paper mentions efforts to reduce time-consuming aspects by adjusting subband usage.
  • Parameter Tuning: The algorithm relies on several parameters (e.g., thresholding values, neighborhood sizes, NSCT levels). Fine-tuning these parameters might be necessary for optimal performance on different types of images.
  • Future Work: The paper implicitly suggests that further research could focus on optimizing computational efficiency and exploring more advanced feature descriptors or matching strategies to handle even more complex scenarios. The authors conclude that the proposed algorithm is robust and accurate, implying its readiness for practical applications.

Summary updated

Abstract

Feature-based registration is an effective and most widely used image registration method currently. It includes three critical steps, feature extraction, feature matching and transformation parameters estimation. This paper mainly explores the first two steps. In one of Chahira Serief’s paper about image registration, feature points extraction based on nonsubsampled contourlet transform (NSCT) was proposed and feature points matching based on Zernike moments was adopted. The registration accuracy and robustness of his algorithm are acceptable, but it can still be improved. In this paper, an improved scheme of this registration algorithm is proposed. The rotation invariance of NSCT-based feature points extraction is improved, which is beneficial to extract homologous feature points. And the reliability and effectiveness of Zernike momentsbased feature points matching are improved, which can improve the matching accuracy. The improved registration algorithm can realize registration of images related by larger scaling, rotation and translation transformation. The simulation results show that the registration robustness is further improved, and the registration accuracy is still high.
BibTeX
@article{ding2016improved,
  author = {Rui Ding and Ziyan Song and Jin Tang},
  title = {{An Improved Image Registration Based on Nonsubsampled Contourlet Transform and Zernike Moments}},
  journal = {International Journal of Signal Processing Image Processing and Pattern Recognition},
  year = {2016},
  doi = {10.14257/ijsip.2016.9.6.17},
}
Figure