When pixels in a two-dimensional matrix are divided into lines A and B in the column (or row) direction (sub-scanning direction) and two neighboring pixels in the row (or column) direction are subjected to gradation correction respectively using characteristic curves for lines A and B, the characteristic curve for line B unpreferably has a stepwise characteristic portion having a step when it is stored in an LUT, due to quantization errors unique to digital values. Hence, the characteristic curve for line A is set to be decreased by the amount corresponding to the step.