PURPOSE: To perform an accurate recognition processing even when two-dimensional symbols on a plane are slantingly read out or when two-dimensional symbols on a cylinder peripheral surface are read out.

CONSTITUTION: Feature points of a two-dimensional symbol image expanded in an image data memory are extracted and a two-dimensional symbol area is detected and segmented according to the extracted feature points. Several feature points are further extracted additionally from the segmented two-dimensional symbol area. Then all the extracted feature points are used as control points to find a function for coordinate conversion from the undistorted image space of the two-dimensional symbols to a distorted image space. On the basis of the coordinate conversion function found as to the two-dimensional symbol image expanded in the image data memory, coordinate positioning in the image data memory is performed to perform image recognition.