Abstract: Linear motion blur estimation is widely studied in the ill-posed deblur related research, but existing methods are too complex, inaccurate or unstable. This study proposes an improved method with two novel contributions to former frequency based method. One is a novel preprocessing step consisting of Hann windowing and histogram equalization. Hann windowing is used to remove the boundary artifacts in the log spectrum while histogram equalization is used to enhance the contrast of the log spectrum. The other is an improved Hough transform based method to estimate the blur direction with cepstrum which eliminates the effect of noise. The method incorporates a least squares line fitting into the traditional Hough transform, and thus accurately estimates the direction. Experimental results demonstrate the efficiency of the proposed method.