A geometric approach to estimate background in astronomical images

TL;DR

This paper introduces a geometric method based on steepest descent to improve background estimation in astronomical images, especially in low-photon count and crowded fields, enhancing faint source detection accuracy.

Contribution

The paper presents a novel geometric algorithm utilizing minima statistics for more accurate background estimation in astronomical images, outperforming traditional methods.

Findings

Achieves background recovery within 10% in low-photon, less crowded images.

Reduces background overestimation to ~14% in crowded fields.

Outperforms conventional methods in accuracy and reliability.

Abstract

Estimating the true background in an astronomical image is fundamental to detecting faint sources. In a typical low-photon count astronomical image, such as in the far and near-ultraviolet wavelength range, conventional methods relying on the 3-sigma clipping and median or mode estimation often fail to capture the true background level accurately. As a consequence, differentiating true sources from noise peaks remains a challenging task. Additionally, in such images, effectively identifying and excluding faint sources during the background estimation process remains crucial, as undetected faint sources could contaminate the background. This results in overestimating the true background and obscuring the detection of very faint sources. To tackle this problem, we introduce a geometric approach based on the method of steepest descent to identify local minima in an astronomical image. The proposed algorithm based on the minima statistics effectively reduces the confusion between sources and background in the image; thereby ensuring a better background estimation and enhancing the reliability of faint source detection. Our algorithm performs well compared to conventional methods in estimating the background even in crowded field images. In low-photon count, less crowded images, our algorithm recovers the background within 10\%, while traditional methods drastically underestimate it by a few orders of magnitude. In crowded fields, the conventional methods overestimates the background by $\sim 200\%$ whereas our algorithm recovers the true background within $\sim 14\%$. We provide a simple prescription to create a background map using our algorithm and discuss its application in large astronomical surveys.