In order to better attenuate random noise of seismic data and get more accurate seismic data reconstruction, the authors, based on the free surface multiples feedback iteration method, employed multidimensional weighted cross-correlation to replace multidimensional weighted convolution, also known as "the focal transformation method". This method is a whole data driven process in which underground information is not required, especially when the local underground geological bodies are complicated and the information that should be considered is large. In order to improve the traditional effective signal based on the focus of the least square calculation transform whose focus is not centrally concentrated, the authors combined 3D curvelet transform and focal transform and used the L1 norm optimization algorithm to get the solution. The combination of 3D curvelet transform with focal transform random noise attenuation of seismic data can make effective signal more concentrated, and the preservation of effective signal becomes more complete after the removal of the noise signal. In comparison with the interpolation method that only uses curvelet transform or focal transform means, the interpolation experiment used in this paper can reconstruct seismic data more completely and sophistically, and the high frequency information can be preserved effectively.