Method of denoising for the image generated by Monte Carlo ray tracing preserving small details

Monte-Carlo methods of ray tracing allow generating physically correct images of virtual scenes, reproducing a large number of optical effects. However, the resulting image contains noise inherent in stochastic algorithms. An original method of denoising in these images is proposed. Small-scale details of the image are not distorted because they are taken from an auxiliary, reference image constructed by a deterministic ray tracing. The reference image allows you to extract the secondary lighting component which is usually a smooth function and allows for efficient filtering. After multiplying its result by an auxiliary deterministic image, a sharp picture is obtained with preservation of small details and with greatly reduced noise.