Abstract: The image processing device finds core linear segments of a linework image by thinning processing; provides a mask area for each end point of the linear segments; integrates the mask areas overlapping with each other into a sum area; with respect to each sum area, finds effective linear segments which have only one end point within each sum area; and judges a feature point in each sum area according to the number of the effective linear segments. Therefore, the device is advantageous in that the feature points are classified more accurately because the linear segments irrelevant to judging the feature points (such as those produced due to distortion and line missing) can be eliminated in the judgment. It is also advantageous in that the data amount to be handled is small and classifying is completed at a high speed since the feature points are classified after vectorization of raster data.