Image prediction method and device

An image prediction method and device. The method comprises: acquiring a first reference unit of an image unit, wherein the image unit and the first reference unit use the same affine model to acquire respective prediction images; acquiring motion information about basic motion compensation units at at least two pre-set positions in the first reference unit; and according to the motion information, acquiring motion information about a basic motion compensation unit of the image unit. Therefore, motion information about the first reference unit of the same affine model is used in multiplexing, which can more accurately acquire a motion vector of the current image unit, maintain the complexity of coding and decoding while improving the prediction accuracy, and improve coding and decoding performance.