PURPOSE: To execute a geometric feature extraction with high accuracy by eliminating the influence exerted by a noise, a deviation, etc., in a jump edge.

CONSTITUTION: A three-dimensional data generating means 1 generates three- dimensional data of a picture element on an image obtained by photographing an object 10, and a weight data generating means 2 generates weight data of each picture element, based on its three-dimensional data. That is, from the three-dimensional data, a picture element of a jump edge is discriminated, and as for the picture elements in the jump edge and its vicinity, and as for other picture elements, small weight data and large weight data are generated, respectively. A geometric feature extracting means 3 extracts a geometric feature of the object 10 on the image by using the three-dimensional data and the weight data. Therefore, to the picture elements in the jump edge and its vicinity subjected to the influence of a noise, etc., small weight data is imparted. Accordingly, in a stage for extracting a geometric feature, the three-dimensional data in the jump edge and its vicinity is scarcely evaluated, and the feature extraction of high accuracy can be executed.