The elastic net method was originally proposed by R.Durbin and D.Willshaw for the traveling salesman problem is now widely applied in different areas of optimization problems. The task of rings recognition in plane is a part of the general image recognition problem. In our case it consists of searching between measured points for groups of points formed rings within a precision of measurements. Ring reconstruction in RICH detectors is complicated by the presence of noise points, overlapping of rings and limited resolution of photon detectors. Moreover, the circle shape of rings can be distorted in a detecting plane of a RICH detector. To solve this problem the elastic neural net is proposed, the internal structure of which can take into account rings deformations and may be easily modified for other similar tasks. An algorithm implementing this method was developed and tested using simulated data for the COMPASS experiment RICH-1 detector at CERN. The efficiency of ring reconstruction was found to be 99%. Description of the algorithm and test results are presented in details.