[Paper Review] Automated detection of gravitational arcs

C. Alard|arXiv (Cornell University)|Jun 30, 2006

Geophysics and Gravity Measurements16 citations

TL;DR

This paper presents a novel method for automated detection of gravitational arcs by computing a local elongation estimator Q(x,y) at each image pixel using second-order moments and marginal distributions after orientation alignment. The approach identifies elongated, narrow structures—characteristic of arcs—via a thresholded Q-map and connectivity-based source reconstruction, achieving high efficiency and accuracy in CFHTLS survey data with a 10x speedup through targeted computation at local maxima.

ABSTRACT

This paper presents a method to identify gravitational arcs or more generally elongated structures in a given image. The method is based on the computation of a local estimator of the elongation. The estimation of the local elongation proceed in two steps: first the local orientation of the structure is computed, then in the next step, a rotation is performed, and the marginal distributions are used to compute the elongation. This procedure allows the computation of the local elongation at each point in the image. Then, using a threshold on the elongation map the elongated structures are identified and re-constructed using connectivity criteria. Finally a catalogue of elongated structures is produced, and the properties of each object are computed, allowing the selection of potential arc candidates. The final selection of the arc candidates is performed by visual inspection of multi-color images of a small number of objects. This method is a general tool that may be applied not only to gravitational arcs, but to all problems related to the mapping and measurement of elongated structures, in an image, or a volume.

Motivation & Objective

• To develop an automated method for detecting gravitational arcs in large-scale astronomical surveys, where manual inspection is infeasible due to data volume.
• To address the challenge of identifying faint, narrow, and highly elongated structures (like gravitational arcs) amidst complex background noise and spurious features.
• To produce a catalog of potential arc candidates by combining a local elongation estimator with connectivity-based source reconstruction and visual inspection for final validation.
• To optimize computational efficiency for large images (e.g., 20000×20000 pixels) by focusing computation on relevant image features such as local maxima.
• To generalize the method beyond gravitational arcs for detecting any narrow, elongated structures in 2D or 3D images.

Proposed method

• At each pixel (x₀,y₀), compute local second-order moments (a, b, c) within a window of size ~2M×2M to estimate the local structure orientation via the angle θ defined by tan(θ) = -c / (-0.5b + 0.5a - 0.5√(b² - 2ab + a² + 4c²)).
• Rotate the local image patch by angle θ to align the major axis with the x-axis, enabling directional analysis along the elongated and perpendicular directions.
• Compute marginal distributions I_X(x) = ∫I(x₀+x,y₀+y)dy and I_Y(y) = ∫I(x₀+x,y₀+y)dx to assess intensity spread along the two axes.
• Apply the elongation estimator Q(x₀,y₀) = (1/(2M)) × I_Y(0) / sup[I_X(x)] over the interval [-M,M], which quantifies local narrowness and elongation.
• Use a threshold on the Q-map to identify candidate elongated structures, followed by connectivity-based source reconstruction to form complete objects.
• Optimize computation by restricting Q(x₀,y₀) evaluation only to local maxima along x and y axes, reducing processing time by ~10× in large images.

Experimental results

Research questions

• RQ1 How can gravitational arcs—characterized by extreme elongation and narrowness—be reliably detected in large astronomical images with minimal human intervention?
• RQ2 What local image estimator can effectively distinguish true arcs from other elongated structures (e.g., spiral arms) or noise artifacts based on shape and width?
• RQ3 How can the computational cost of computing a local elongation estimator across a large image be minimized without sacrificing detection accuracy?
• RQ4 To what extent can the proposed estimator Q(x,y) differentiate between arclets and other narrow, elongated features using measurable morphological parameters?
• RQ5 Can the method be generalized to detect any narrow, elongated structure beyond gravitational arcs in imaging data?

Key findings

• The estimator Q(x,y) reaches values up to ~3 within the central arclet in the test image, significantly higher than in surrounding regions, confirming its sensitivity to narrow, elongated features.
• The method successfully isolates the central arclet and other elongated structures from background noise and spurious features using Q-map thresholding and connectivity criteria.
• The 5 largest structures detected in a 1000×1000 pixel region had mean Q-values between 1.59 and 2.19, with the central arclet having 103 pixels and a radius of 26.8 pixels.
• A small, faint arclet near the lensing center was clearly detected in the Q-map (Fig. 4), demonstrating sensitivity to low-surface-brightness features.
• The method reduces computation time by a factor of ~10 by restricting Q(x,y) evaluation only to local maxima along x and y axes, enabling scalability to large surveys.
• The optimal filter scale σ for maximizing signal-to-noise in the estimator is σ ≈ 1.19a, where a is the PSF width, ensuring optimal response to narrow structures.