AbstractAlthough the Hough transformation is a powerful method to detect segments in an image, it requires not only a vast amount of memory but also a long processing time when a highly accurate detection is demanded. This paper proposes a new method of detecting segments with high accuracy using the Hough transformation without increasing the amount of memory. Unlike the conventional segment detection method in which the detection process is based on the parameter plane, this new method feeds the process back to the image plane from the parameter plane. Another feature of the method is the elimination in the parameter plane of Hough curves which are derived from pixels in the image plane constituting already detected segments. Experiments show that this method detects segments with a high accuracy from “noisy” images including many long and short segme