Einstein rings are rare gem of the strong lensing phenomena. Unlike doubly or quadrouply lensed systems which only provide lens mass profile constraints at 1 or 2 position angles, the ring images can be used to probe the underlying lens gravitational potential at every single position angles, thus enable us to put much tighter constraints on the lens mass profile. In addition, the magnified and stretched images of the background source also enable us to probe properties of high-z galaxies with enhanced spatial resolution and higher signal to noise ratios, which is otherwise not possible for un-lensed galaxy studies. Despite their usefulness, only a handful of Einstein rings have been repoted up-to-date, mainly based on serendipitous discoveries or visual inspections of hundred thousands of massive galaxies -i.e. luminous red galaxies (LRG) -or galaxy clusters -i.e. brightest cluster galaxy (BCG). In the era of large sky survey, and with the upcoming surveys such as Large Synoptic Survey Telescope, visual inspection to discovery Einstein rings is very difficult, if not impossible, and an automated approach to identify ring pattern in the big data to come is in high demand. Here we present an Einstein ring recognition approach based on computer vision techniques. The workhorse of this approach is the circle Hough transform, which can recognise circular patterns or arcs at any given position with any radius in the images. We devise a two-tier approach: first pre-select LRGs associated with multiple blue objects as possible lens galaxies, than feed these possible lenses to Hough transform to identify Einstein rings and arcs. As a proof-of-concept, we investigate our apprach using the Sloan Digital Sky Surveys. Our results show a 100% completness, albeit a low purity at 40%. We also apply our approach to three newly discovered Einstein rings and arcs, in the Dark Energy Survey, Hyper Suprime-Cam Subaru Strategic Program, and UltraVISTA survey, illusting the versatiliy of our approach to on-going and up-coming large sky surveys in general. The beauty of our approach is that it is solely based on JPEG images, which can be easily obtained in batch mode from SDSS finding chart tools, without any pre-processing of the image. An implementation in Python will be available and can be downloaded from the author.