The beamforming algorithm obtains the azimuth estimation of the target by extracting the information from the data covariance matrix. Many studies have shown muchunextracted information in the covariance matrix. This paper proposes the Estimate Parameter via Projecting a Subspace onto a Dictionary(EPPSD) algorithm that may further extract information from the covariance matrix. Through the regularization constraint, we limit the spatial power distribution to the signal subspace to ensure the correctness of the target estimation. In addition, we use the weighted 1-norm to constrain the spatial power distribution vector to ensure that the targets in the space remain sparse. In the case of the above two constraints, the beamforming results are fitted by a dictionary, and the azimuth estimation of the target is obtained by quadratic programming. The experimental results show that the algorithm can estimate the target azimuth despite the inaccurate sparse constraint.