To make it possible to execute proper image processing independently of an image exposure state by properly extracting a reference density value to be utilized for the determiniation of arithmetic processing for executing image processing.

An exposure state of an image of an original G is found out by judgement based on a neural network. Then a non-effective gradation area NE of a prescribed range (e.g. a range of 0 to 10% of a density value) is determined from the found exposure state and a range excluding the non-effective gradation area NE is determined as an effective density area EE. Then the density value D1 of a high light point is extracted as a reference density value from density values included in the area EE. Then a density conversion LUT is prepared based on the accumulated frequency of plural density values including the density value D1.