An optimized method is proposed to estimate the turbulent point spread functions （PSF） by using two consecutive short-exposure images in a sequence of turbulence-degraded images, in which the principle of constraint optimization has been applied in the process of restoration. The scheme of preserving peaks of the PSF has been adopted in the estimation of the PSF values, which is in agreement with the property of turbulent PSFs being stochastic changeable with some complex peaks. The problem of calculating the PSFs＇values has been boiled down to optimization estimation based on relax iterations. The discrete values of the PSFs can be estimated by minimizing the critical function, consequently the object image can be obtained. The restoration experiments show that the proposed algorithm is efficient with good restoration effect and fast speed.