Orthogonal pulse pairs generated by the polarization beam splitter (PBS) and the polarization maintaining-optical switch (PM-PSW) can effectively suppress the polarization fading in phase-sensitive optical time-domain reflectometry (Φ-OTDR) systems, but the PM-PSW also brings a lot of noise when switching the optical path periodically. Therefore, a non-local means (NLM) image-processing method is proposed to enhance the signal-to-noise ratio (SNR) of a Φ-OTDR system. Compared with the existing traditional noise reduction methods based on the one-dimensional signal, the method makes full use of redundant texture and self-similarity of multidimensional data. The NLM algorithm can obtain the estimated denoising result value of current pixels by the weighted average of pixels with similar neighborhood structures in the Rayleigh temporal-spatial image. To validate the effectiveness of the proposed approach, we have carried out experiments on the actual signals obtained from the Φ-OTDR system. In the experiment, a sinusoidal waveform of 100 Hz is applied at 20.04 km of the optical fiber as a simulated vibration signal. The switching frequency of PM-PSW is set to 30 Hz. The experimental result shows that the SNR of vibration positioning curve is 17.72 dB before denoising. After using the NLM method based on image-processing technology, the SNR reaches 23.39 dB. Experimental results demonstrate that this method is feasible and effective in improving SNR. This will help to realize accurate vibration location and recovery in practical applications.