Studies On Blind Deconvolution Of Adaptive Optical Images

For large-aperture ground-based telescopes, the space target imaging quality tends to be seriously degraded due to atmospheric turbulence. The adaptive optics system can greatly improve the image quality by real-time detecting and correcting wave-front distortions caused by atmospheric turbulence. However, the correction is usually incomplete due to limited closed-loop bandwidth and the wave-front sensing noise, etc. To better take advantage of adaptive optical images, post-processing techniques must be applied to improve image quality. The blind deconvolution is an important image restoration technique of adaptive optical images. Based on optical imaging principles, the blind deconvolution employs strict statistical estimation theories for the inversion of optical images, and the restored image can approach the diffraction limit of the optical system if the numerical model is accurate enough. Based on previous studies on blind deconvolution algorithm of adaptive optical images, this paper improves the algorithm in the following two aspects:1) The implementation method of regularization constraints on the previous algorithm is studied, which solves the problems of non-stable convergence and over-iteration caused by insufficient constraints. The Tikhonov regularization method and the total variation(Total variation, TV) regularization method are studied, and their respective effect on the algorithm is discussed. In both case the modified algorithm showing significant improvement in stability and noise suppression.2) The method of using Zernike polynomials to express the point spread function is studied, based on which two methods of initial-value selecting are proposed and compared. The implementation of Zernike polynomials is realized, and the algorithm convergence speed is improved. Four methods of initial-value selecting are introduced and demonstrated with simulated and real blurred images. Results show that better restoration appears when applying Kolmogorov-spectrum method and Gaussian PSF-fitting method.