PURPOSE: To prevent visual deterioration in a picture by obtaining an error between a picture element of the original picture and a predicted value obtained by the picture element and setting a step width of inverse quantization and quantization in response to the error variably so as to avoid concentration of a portion with a large coding error in one picture.

CONSTITUTION: An error evaluation means 9 receives a picture element of an original picture stored in a 1st storage means and a predicted value corresponding to an original picture obtained from the photoelectric conversion element subject to 2-dimension discrete cosine transformation, quantization and inverse quantization and 2-dimension discrete inverse cosine transformation and compares them to obtain a coding error, and outputs a control signal used to revise a quantization step width of a quantization means 3 and an inverse quantization step width of an inverse quantization means 5 depending on the coding error. That is, when the error is large, the visual deterioration of picture quality is prevented by setting the step width of quantization and inverse quantization to be small.