PURPOSE: To attain data compression with strict reversible characteristic by assigning a longer code than an original data length when the absolute value of a predictive error is shorter than a value and a longer code than the original data length when the absolute value is larger than the value.

CONSTITUTION: A delay circuit 1 retards an input picture element data by one picture element. A circuit 2 multiplying a predictive coefficient multiplies (a) with a data from the circuit 1, where (a) is the predictive coefficient. A subtracter 3 calculates a difference between a current data and a predictive value by the circuits 1, 2 to obtain a predictive error (e). A controller 4 discriminates the error (e) and operates the coder 5 in the (m)-bit output mode in the case of the relation of -(2m-1-1)≤e≤(2m-1-1) and operates the coder 5 in the (m+n)-bit output mode ithe case of the relation of e<(2m-1-1) or e>(2m-1-1). A parallel serial converter 5 converts the m-bit code or the (m+n)- bit code outputted from the coder 5 into a serial signal and outputs it. Moreover, the control section 4 applies input data control and inputs the data matching the parallel/serial conversion.