Abstract: A method of segmenting the field of motion of an image, particularly a video image, including forming a blockwise field of motion from motion vectors allocated to image pixels, processing the blockwise field of vectors in order to extract from this field a set of dominant vectors of motion, by elimination of the vectors close to the dominant vectors, so as to retain only representatives of widened classes of motion. Then a vector of the reduced set of dominant vectors is reassigned to each block, and the resulting field, after time filtering, is used for the encoding of the image by "quad-tree" encoding. This disclosed method applies in particular, to the setting-up of the coding-auxiliary digital data to the limit of 1 Mbt/s.