A Novel Strategy for Random Noise Attenuation Using Adaptive Frequency Non Local Means

NLM (Non local means) is a powerful method for attenuating seismic random noise. It utilizes the redundant information on similar patches to enhance the data quality. However, due to the large number of the patches in seismic sections, this method is too slow to be applied in the real data processing. Besides, there will be some signal leakage, sometimes, when applying it to the data with complicated structure. In this paper, we propose a novel strategy called AFNLM (adaptive frequency non-local means) that implements the NLM in frequency-space domain which will tremendously improve the speed of traditional NLM method. We also establish an experiential equation between the data spectrum and the weighted parameters to adaptively adjust the severity of the filter, which will help to produce a better denoised result when the structure of seismic data is not ordinary. The synthetic and real examples are provided to prove the validity of this method.