The quality of motion-blurred image restoration is highly dependent on the blur direction identification, which is an important and difficult problem in image restoration. In frequency domain, we summarize four important characteristics of motion-blurred image, and then propose a method to reduce the impact of deficiencies in frequency characteristics by using the binaryzation operation, the morphological operations and the main-connectivity detection method. Aiming at identifying the blur direction effectually, we propose an algorithm to detect the medial axis of white stripes by combining three second order moments of motion-blurred image in frequency domain. The algorithm in this paper is quite different from traditional algorithms which detect black stripes. Experimental results show that our algorithm is much faster and can achieve high accuracy and stability.