METHOD FOR CORRECTION OF EYE IMAGE USING MACHINE LEARNING AND METHOD OF MACHINE LEARNING Russian patent published in 2016 - IPC G06T1/00 G06T5/50 H04N5/00 

FIELD: medical equipment.

SUBSTANCE: group of inventions relates to medical equipment, in particular, to means of processing images and video image of eye interlocutors during video chat, video conferencing. Method of machine training a precursor to correct orientation of sight on an image comprises obtaining a plurality of pairs of images, containing inside each pair of image of same person, determining position of eyes on each pair of images, training predictor, generating a correcting displacement vector, so that for each pair of images with replacement of colour components of each pixel of first image from pair with colour components of another pixel of first image from pair, offset according to prediction of predictor, an image is obtained, maximally similar to second image of pair and storing predictor. Method for correction of image of eyes is characterised by that it comprises loading a predictor, obtaining at least one frame of person's face, determining position of person's eye on image and forming two rectangular areas, closely described around eyes, replacing colour components of each pixel in area of eyes with colour components of pixel offset according to prediction of predictor of machine training.

EFFECT: technical result of invention is improvement of accuracy of correction of eye image while reducing resource consumption of video processing process.

13 cl, 4 dwg