In the disclosed predictive image coding method, multiple sets of motion information are stored in a motion information recording means, and one or more pairs comprising two of said sets of motion information having different structural element values are derived. Using the two sets of motion information of a pair selected from said one or more pairs, a predictive signal of a target region in an image is generated by means of motion compensation. The two sets of motion information of said selected pair are stored in the motion information recording means for generation of predictive signals of other regions.
展开▼