This paper presents a new exact method for the detection of edges in digital grey level imagery. This method performs expanding and compressing algorithms on the binary graph of classification of the original image in order to delete the additive white Gaussian noise and other disturbances

and to extract a coarse boundary of object. Then

within the neighborhood of this boundary

we trace the edge pixels based on an isotropic gradient operator and orientation information. Furthermore

we estimate the object’s proportion of each edge pixel

using a local optimal proportion estimation operator （so called subpixels segmentation）. Combining the proportion values and forword orientation of edge

we compute intersections of boundary curve and each edge pixel. After geometric correction of all intersections

we output the smooth curve of object edge and its area

using automatic computer graphic techniques. Finally

through experimental analysis of analog data and actual image data by using above process

we show that this method provides us an optically better edge than any others which output an edge by taking pixel as a unit

that it also gives us a more accurate position of boundary and area of object. This has a quite significance for combining automatic graph techniques with image process techniques

automatic mapping of remote sensing data

large scale mapping of low resolution data image

and for application fields which demand sub-pixel accuracy area survey of objects.