Optimal and Efficient Streak Detection in Astronomical Images

• Published in: The Astronomical journal Vol. 156; no. 5; pp. 229 - 241
• Main Authors: Nir, Guy, Zackay, Barak, Ofek, Eran O.
• Format: Journal Article
• Language: English
• Published: Madison The American Astronomical Society 01.11.2018; IOP Publishing
• Subjects: Algorithms; Asteroid detection; Asteroids; Astronomy; Bleeding; Celestial bodies; Computer simulation; Cosmic rays; Global positioning systems; GPS; Image detection; methods: data analysis; Near-Earth Objects; Pixels; Point spread functions; Radon; Radon transformation; Satellite imagery; Satellites; Signal to noise ratio; techniques: image processing
• ISSN: 0004-6256; 1538-3881
• DOI: 10.3847/1538-3881/aaddff

Identification of linear features (streaks) in astronomical images is important for several reasons, including: detecting fast-moving near-Earth asteroids; detecting or flagging faint satellites streaks; and flagging or removing diffraction spikes, pixel bleeding, line-like cosmic rays and bad-pixel features. Here we discuss an efficient and optimal algorithm for the detection of such streaks. The optimal method to detect streaks in astronomical images is by cross-correlating the image with a template of a line broadened by the point-spread function of the system. To do so efficiently, the cross-correlation of the streak position and angle is performed using the Radon transform, which is the integral of pixel values along all possible lines through an image. A fast version of the Radon transform exists, which we here extend to efficiently detect arbitrarily short lines. While the brute force Radon transform requires operations for a N × N image, the fast Radon transform has a complexity of . We apply this method to simulated images, recovering the theoretical signal-to-noise ratio, and to real images, finding long streaks of low-Earth-orbit satellites and shorter streaks of Global Positioning System satellites. We detect streaks that are barely visible to the eye, out of hundreds of images, without a-priori knowledge of the streaks' positions or angles. We provide implementation of this algorithm in Python and MATLAB.