An image coding method is utilized to provide high quality images without error
accumulation. Such an image coding method comprises estimating motion vectors between
an input image to be coded and a reference image; synthesizing a prediction image
of the input image using the motion vectors and the reference image; generating
a difference image by calculating a difference between the input image and the
prediction image; and outputting coded information of the input image including
information related to the difference image and the motion vectors. The prediction
image is synthesized by calculating intensity values at points where no pixels
actually exist in the reference image by interpolation. Such an interpolation is
performed according to information specifying a positive rounding method or a negative
rounding method when the current frame is a P frame, and using a predetermined
rounding method which is a positive rounding method or a negative rounding method
when the current frame is a B frame.
