Objective The surface images of nano-scale thin film samples can be obtained by the scanning probe microscope device represented by atomic force microscope, but these images have different degrees of noise, which affect the quality of the image and the judgment of image information. In order to obtain the surface state of the film more accurately, noise reduction of the surface image data and information of the sample is needed. Methods This paper proposed a multi-layer wavelet decomposition noise reduction algorithm based on the introduction of the basic theory of wavelet transform and the analysis of noise sources by combining the imaging characteristics of equipment such as AFM and the time-frequency locality characteristics of wavelet transform. Fourier transform is the basis of the traditional theory of signal, but it has some limitations as a kind of global change. Fourier transform fails to simultaneously describe the local characteristics of the time domain and the frequency domain, which are the key parts of unstable signal characteristics. The wavelet transform kept the advantage of window Fourier transforms in localization and changed its defect of fixed size. Results The frequency of the original image signal was distributed in the range of 0~4000 Hz. After the wavelet transform, the signal waveform was more smooth, and the frequency spectrum was distributed between 500 Hz and 2000 Hz. Conclusion The wavelet transform was applied to signal noise reduction of the film surface images, and the experiment proved that the noise could be effectively removed using wavelet transform.