A descriton of an image involves properties of the image or of its parts, and relationships among the parts. There are many methods of describing images in terms of properties and relationships. Here a description via feature point is discussed, and this technique is applied to Korean letter images. The extraction of feature point, in general, is performed on the image after thinning it. In this paper, the feature point is get from the original image directly rather than from the thinned one. In other word the letter width do not affect the extraction of feature point. The image plane is scanned with an interval which depends on the width of given letter. The eight neighbor elements of an element which is defined as width by width pixels are checked for obtainning its orientation and determining its property, that is, whether this element has feature point or not. The letter image can be described as its critical points (property) and their orientation (relationships). The Korean letter image can also represented in its feature points and their directions. Some strictly-written data are examined with over 90\% correctly.