PURPOSE: To improve the prediction efficiency and encoding efficiency by suppressing a quantization noise by performing control so that a pass band is narrower as quantization step size is larger

CONSTITUTION: A difference circuit 202 calculates the difference of an input block from a prediction signal to generate a prediction error signal, which is encoded and quantized by a DCT circuit 204, whose output is sent to a reception side from a terminal 218. This data is decoded by a decoding circuit 209 inversely to the circuit 204. A variable characteristic filter 212 is used to remove the quantization noise in the prediction signal and step size information on the quantization of respective blocks is read out of a step size memory 211. As for this information and a prediction block position indicated by a motion vector detecting circuit 213, control is performed according to the frequency of the power spectrum of the prediction signal of a power spectrum estimating circuit. Then the pass band becomes narrow as the quantization step size becomes large, and no quantization noise is included, so the prediction efficiency and encoding efficiency are improved.