An image processing device includes a feature emphasis unit for extracting first image frequency information from image data for each first unit area, a boundary provisional determination unit for defining a value obtained by adding a predetermined weight to the first image frequency information as representative feature information, and provisionally determining as a boundary a first unit area whose variance from the representative feature information of the adjacent area is at or higher than a predetermined level, and a boundary determination unit for extracting the second image frequency information for each second unit area smaller than the first unit area in the range of a provisionally determined position and the vicinity, and generating boundary information using as the boundary a second unit area whose value based on the variance from the second image frequency information of the adjacent area is at or higher than a predetermined level.