Hough transform, which is widely being used to detect geometric features like lines in an image, has a problem of high processing time cost. In this paper, we propose an improved Hough transform using directional information in an image for the problem. Pixels of the image have their own directional information and magnitude. Therefore, we can predict the directions of lines in the image by statistically analyzing the directional information of pixels, and improve the problem of Hough transform from the predicted directions. Our proposed method could improve about 50% and 59% of the processing time in comparison with previous Duda method and Fernandes method, respectively, and showed a good performance in a part of lines detection.