SAR images generated on the basis of FFT suffer the poor resolution. In super-resolution algorithms of SAR/ISAR image the modern spectral estimation technique is usually used, such as Minimum Variance Method (MVM), AR model, eigen-vector, MUSIC and maximum entropy, to improve the resolution of the image. High-resolution spectral estimation of finite length sample is considered an underdetermined problem. In the framework of Bayesian criteria, a prior probability density function (pdf) of the spectra is included as a regular term in the cost function for MAP estimation. To improve the efficiency of calculation in 2-D case, fast algorithm is derived. The better resolution is achieved by the method while the method is applied to SAR image peak feature extraction.