Abstract：In data processing,conventional SVD filter fails to nicely raise the signalJnoise ratio of non-horizontal events. By using the statistical characteristics of the second rank gray level of the quantified seismic irnage,we derive the major event strikes in a given processing area from the energy,deviation and homogeneity characteristics which have been obtained in the associated matrix. Then the input matrix of SVD filter can be adjusted automatically,and the characteristics and singular-value distribution curve are used to divide the given processing area into some different parts and to determine how many singular values the SVD filter will take,thus improving the performance of SVD filter. In processing synthetic and real seismic data,the irnproved SVD filter brings better results than conventional SVD filter does.
