A denoising and enhancing method based on frequence domain filter was developed aiming at two main noise in cement pavement image: groove noise and longitudial stripe noise. Firstly, the spectral features of the two kinds of noise were analyzed. On this basis, the position of groove spectral peaks in the frequency domain were deduced according to the groove period in spatial domain. Then, a series offilters were designed to elimiante these peaks in the frequency domain. Moreover, aiming at the longitudianl stripe noise belonging to the low frequency components, a filter of which parameters were optimized through experiments, was designed to eliminating this disturbance by suppressing low frequency and enhancing high frequency. Finally, the experimental results showed that this proposed method can eliminate the two kinds of noise and enhance the constrast of cracks, which provided a foundation for crack recognition in cement pavement.