Single-shot non-invasive imaging through a scattering medium has been an area of research focus for several years. Achieving an accurate autocorrelation distribution is crucial for effective image reconstruction. However, the obtained autocorrelation is always contaminated by various types of noise. In this study, we propose an effective method to extract the autocorrelation distribution of the sample from the noise-laden raw data before image retrieval. This is accomplished by incorporating an intensity threshold and a spatial Gaussian filter into the retrieval algorithm, which effectively eliminates the noise artifacts in the reconstructed image. The proposed algorithm exhibited remarkable robustness for both visible (laser, white light, and light-emitting diode light) and X-ray light sources, and its efficacy was verified through multiple experiments.