The disclosure relates to a method for predicting a current block of pixels in a current frame, comprising: - obtaining (11) a first prediction of the current block, delivering a first predicted block, - updating said first predicted block by implementing at least one iteration (12) of the following steps: obtaining (121) a current patch formed by a current template of the current block and an updated predicted block obtained at a previous iteration; o searching (122) for K nearest neighbor patches of said current patch in a decoded part of the current frame or in at least one reference frame; o determining (123) a linear combination of the K nearest neighbor patches which approximates the current patch, or of the K templates of the nearest neighbor patches which approximates the current template, according to a predetermined criterion, obtaining (124) an updated predicted block by applying said linear combination to the corresponding K blocks of the nearest neighbor patches.