The invention relates to a digital symmetrical azimuth tensor filtering method, which is technically characterized in that: digital signals express local information as symmetrical tensor by pretreatment (1); the symmetrical tensor obtains local azimuth information corresponding to each local position by azimuth generation (2); the local azimuth information controls and generates symmetrical azimuth filtering kernel (3); the symmetrical tensor obtains a new symmetrical tensor by tensor convolution filtering (4). The method of the invention has two outstanding advantages: one is that the localstructure of the signals is expressed as the symmetrical tensor, which can express the local structure of the signals more finely and accurately and reduce information loss in the follow-up processing; the other is that symmetrical azimuth tensor filtering is carried out to the symmetrical tensor to lead to the fact that the processing to the tensor is more in accordance with the discrimination power of the sense organ of human being, thus having stronger noise-proof capacity.