To consider the amount of information of a motion vector generation, to quantize a predicted error as precisely as possible and to produce high image quality with a ultra low bit rate by minimizing a quantization step size of the predicted error.
An orthogonal transformation circuit 7 prepares an output for indicating a function f(SAD, Q) making an orthogonal transformation SADDCT for orthogonal- transforming motion predicted error information outputted from a subtraction circuit 6 equivalent to an information generation amount predicted value at an information generation amount prediction circuit 8. Also, a difference vector which is equivalent to the difference between a motion vector from a motion vector generation circuit 4 and a motion vector of a motion vector memory 9 is fetched from a subtraction circuit 10, performs a step-size prediction at a step-size prediction circuit 11 and a difference vector selected at a minimum value discrimination circuit 12 is encoded at an encoding circuit 16, so that the difference vector becomes minimum. A motion vector outputted from the motion vector memory 9 is added to an output of the minimum value discriminating circuit 12 at an addition circuit 13, and a required delay is given at a delay circuit 14 and stored in the motion vector memory 9 as an updated motion vector.
MATSUMOTO SHUICHI