Method of compressing a digital image signal in which a first quantization step set, which is unique for a given segment, is determined so that the number of bits needed to encode the quantized data corresponding to this segment is greater than a target value. This first quantization step set then being modified, as a priority, for the blocks of the segment for which the gain, in the course of this modification, on the reduction of the number of bits needed to encode the quantized data corresponding to the segment to which it belongs, is the highest. This modification is carried out, on as many blocks as is necessary for the number of bits of this segment to be less than or equal to the target value. Device to implement this method.