Comparison of Five Numerical Codes for Automated Tracing of Coronal Loops

The three-dimensional (3D) modeling of coronal loops and filaments requires algorithms that automatically trace curvilinear features in solar EUV or soft X-ray images. We compare five existing algorithms that have been developed and customized to trace curvilinear features in solar images: i) the oriented-connectivity method (OCM), which is an extension of the Strous pixel-labeling algorithm (developed by Lee, Newman, and Gary); ii) the dynamic aperture-based loop-segmentation method (developed by Lee, Newman, and Gary); iii) unbiased detection of curvilinear structures (developed by Steger, Raghupathy, and Smith); iv) the oriented-direction method (developed by Aschwanden); and v) ridge detection by automated scaling (developed by Inhester). We test the five existing numerical codes with a TRACE image that shows a bipolar active region and contains over 100 discernable loops. We evaluate the performance of the five codes by comparing the cumulative distribution of loop lengths, the median and maximum loop length, the completeness or detection efficiency, the accuracy, and flux sensitivity. These algorithms are useful for the reconstruction of the 3D geometry of coronal loops from stereoscopic observations with the STEREO spacecraft, or for quantitative comparisons of observed EUV loop geometries with (nonlinear force-free) magnetic field extrapolation models.