RingFinder: automated detection of galaxy-scale gravitational lenses in ground-based multi-filter imaging data

We present RingFinder, a tool for finding galaxy-scale strong gravitational lenses in multiband imaging data. By construction, the method is sensitive to configurations involving a massive foreground early-type galaxy and a faint, background, blue source. RingFinder detects the presence of blue residuals embedded in an otherwise smooth red light distribution by difference imaging in two bands. The method is automated for efficient application to current and future surveys, having originally been designed for the 150-deg2 Canada France Hawaii Telescope Legacy Survey (CFHTLS). We describe each of the steps of RingFinder. We then carry out extensive simulations to assess completeness and purity. For sources with magnification mu>4, RingFinder reaches 42% (resp. 25%) completeness and 29% (resp. 86%) purity before (resp. after) visual inspection. The completeness of RingFinder is substantially improved in the particular range of Einstein radii 0.8 < REin < 2. and lensed images brighter than g = 22.5, where it can be as high as 70%. RingFinder does not introduce any significant bias in the source or deflector population. We conclude by presenting the final catalog of RingFinder CFHTLS galaxy-scale strong lens candidates. Additional information obtained with Hubble Space Telescope and Keck Adaptive Optics high resolution imaging, and with Keck and Very Large Telescope spectroscopy, is used to assess the validity of our classification, and measure the redshift of the foreground and the background objects. From an initial sample of 640,000 early type galaxies, RingFinder returns 2500 candidates, which we further reduce by visual inspection to 330 candidates. We confirm 33 new gravitational lenses from the main sample of candidates, plus an additional 16 systems taken from earlier versions of RingFinder. First applications are presented in the SL2S galaxy-scale Lens Sample paper series.