Description
Source extraction in extragalactic HI radio surveys has traditionally made extensive use of visual, by-eye inspection. Efforts to quantify its completeness and reliability have, however, been limited by the often-laborious nature of the procedure. The result is that there are many unanswered questions and persistent misconceptions : is visual source finding really totally subjective, or can its recovery rates be reliably and predictably quantified ? Can it find fainter sources than algorithmic methods, or are modern extractors able to out-perform human source-finding in depth as well as speed ? I will present the results of a large experiment designed to address these issues. From a galaxy-free data cube, I extracted small subset "cubelets" into which artificial sources were injected to mimic the properties of spatially unresolved HI emission. A dedicated extraction tool was used to maximise the search speed so that about 800 cubelets could be searched per day. I searched 8,500 such cubelets containing a 4,232 sources, with line widths and peak S/N varying to mimic typical galaxy properties. The results fully sample the parameter space of completeness and reliability, from sources which were completely undetectable by eye to those which were detected with 100% accuracy. I show that a striking trend emerges between the completeness and integrated S/N levels, with visual extraction offering substantial gains over algorithmic extractors. Visual source finding can also be rapid enough to allow searches even of the large data sets expected with the SKA and its precursors.
