A fast photometric image alignment algorithm with row and column means

Jie Zheng,Linqiao Jiang,Jianfeng Tian

DOI: https://doi.org/10.1016/j.newast.2024.102224

2024-04-16

Abstract:This paper introduces an astronomical image alignment algorithm. This algorithm uses the means of the rows and columns of the original image for alignment, and finds the optimal offset corresponding to the maximum similarity by comparing different offsets between images. The similarity is evaluated by the standard deviation of the quotient divided by the means. This paper also discusses the theoretical feasibility of this algorithm. Through practical testing, it has been confirmed that the algorithm is fast and robust.

Instrumentation and Methods for Astrophysics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the relative displacement problem in astronomical image alignment. Specifically, the author proposes a fast photometric image alignment algorithm based on row and column means, aiming to solve the relative displacement problem of astronomical images during the observation process caused by telescope movement or other factors. This problem is particularly important when dealing with the light - curve of specific targets (such as binary stars, exoplanets and supernovae), because these studies rely on the consistency of the relative positions between images.

### Problem Background

In astronomical observations, especially in the processing of time - domain observation data, the absolute celestial coordinates are not always important, while the relative displacement between images is crucial. Many existing image alignment tools (such as astroalign, pattern - matching algorithms, etc.) rely on star catalogs extracted from images, which not only increases the computational complexity but also requires the invocation of external programs or tools. In addition, these methods usually need to perform bias correction and flat - field correction, increasing the processing time and complexity.

### Proposed Solution

To solve these problems, this paper proposes a new image alignment algorithm with the following characteristics:

1. **Does not rely on star catalogs**: This algorithm directly uses the row and column means of the image for alignment without extracting star catalogs from the image.
2. **No need for correction**: The algorithm can run on data without bias correction and flat - field correction, simplifying the processing flow.
3. **Fast and robust**: By calculating the similarity (based on standard deviation) under different offsets, the optimal offset is found to ensure that the alignment process is both fast and reliable.

### Algorithm Principle

The main steps of this algorithm are as follows:

1. **Calculate row and column means**: Calculate the row means ($\text{mean}_y$) and column means ($\text{mean}_x$) of the reference image and the target image respectively.
2. **Calculate similarity**: Evaluate the similarity by comparing the ratios of the row and column means under different offsets and calculating the standard deviation of these ratios. The smaller the standard deviation, the higher the similarity between the images.
3. **Determine the best offset**: Select the offset with the smallest standard deviation as the final alignment offset.

The formulas are expressed as follows:
- Row mean: $\text{mean}_y=\frac{1}{N}\sum_{i = 1}^{N}I(i,j)$
- Column mean: $\text{mean}_x=\frac{1}{M}\sum_{j = 1}^{M}I(i,j)$
- Similarity evaluation: $\sigma=\sqrt{\frac{1}{N}\sum_{i = 1}^{N}\left(\frac{\text{mean}_{yt}(i+\delta)}{\text{mean}_{yr}(i)}-1\right)^2}$

where $\delta$ is the offset, and $\text{mean}_{yt}$ and $\text{mean}_{yr}$ are the row or column means of the target image and the reference image respectively.

### Experimental Verification

Through practical tests, this algorithm has been proven to be fast and robust, and can maintain high precision and efficiency when processing a large number of astronomical images. The experimental results show that this algorithm performs excellently when processing the observation data of the LP UMa binary star, and can accurately align more than 1,000 images.

### Conclusion

The fast photometric image alignment algorithm based on row and column means proposed in this paper is theoretically feasible and shows fast, reliable and robust characteristics in practical operation. This algorithm is suitable for image alignment tasks with only translation, but not suitable for processing images with rotation or different pixel scales.