Method for producing super-resolution images and nonlinear digital filter for implementing same

A method of image super-resolution comprising of:

• exposure of plurality of frames, said plurality of frames read-out from a sensor, alignment of said plurality of frames, enlarged image formation, enlarged image filtration,
• wherein the plurality of frames are read from the digital sensor in the form of a continuous sequence of high frame rate frames, the enlarged image is formed by combining input frames of low resolution, and resolution enhancement is achieved by applying the nonlinear filter to the enlarged image.

A nonlinear digital filter for image super-resolution, input data of which are pixels of the image to which said super-resolution processing is applied, said filter further includes the pretrained with the help of reference image neural network, wherein modified digitized pixel data are transferred to the neural network, said modification includes:

• low-frequency component selection, element-wise pixels grouping, low-frequency component subtraction from grouped pixels followed by data normalization, said neural network output data are being inversely normalized, and said low-frequency component is added to the value of the neural network output to produce final output of a filter.