Abstract:With technology development inSeismicdata processing and interpretation, automatic interpre-tation of seismic data became possible.However,automatic interpretation of seismic data requireshigher signal noise ratio(SNR).In order to im-prove the SNR of seismic data and preserve faultand other information, we propose here a new seis-mic image processing method based on anisotropicdiffusion. This approach has three ingredients:firstly,computing the structure tensor which pro-viding the information on local orientation of thereflection layers;then, applying the seismic coher-ence attribution in the diffusion tensor as the faultspreserving factor;and finally, smoothing along thereflection layers and preserving the faults. Bothmodel and real data examples show that this meth-od obtains remarkable improvements.