In order to resolve the problem of light closed loop for the level of fine tracking system of survey camera high-precision image stabilization control, a star location method is proposed based on nonsubsampled contourlet transform (NSCT) and mapped least squares support vector machine (MLSSVM). Aiming at the characteristics of the star image, the image is denoised by adaptive NSCT. By analyzing the systematic errors of square centroid method in the frequency domain, its approximate analytic expression is obtained. By using Monte-Carlo numerical simulation method, regression analysis based on MLSSVM with radial basis function (RBF) kernel is proposed. The nonlinear function between the ideal star centroid location and the systematic errors is obtained, and is used to correct the systematic errors. Simulation results show that the proposed method is stronger in anti-noise performance and the star location accuracy is improved by 1 to 2 order of magnetude.