To obtain high resolution human retinal images, an adaptive optical system for retinal imaging was established. In the system,the residual aberration of the imaging system was used as initial parameter estimation for image restoration to get high resolution images by semi-blind deconvolation.First,the adaptive optical system consisting of a Hartmann-Shark wavefront sensor and a micromachined membrane deformable mirror was use to correct the dynamic human eye wavefront aberrations and to access the residual aberration of system when retinal images were capturing. Then,the optical transfer function was used as initial parameter estimation of image deconvolution modal to perform the iterative semi-blind constraint deconvolution on the retinal image and to eliminate the influence of residual aberration on imaging quality and obtain the high resolution retinal images. The experiment results show that satisfactory retinal image can be gotten by the proposed method. The image quality has been improved nearly once, and the success rate of imaging has raised from 38% to 78%. Meanwhile, the correction time reduces by 6/7. It concludes that the calibration time is reduced effectively, the success rate of imaging is improved, and the the scope of application is expanded by proposed method.