Objective The current image deblurring removal mechanism ignores the image space structure characteristics, which reduces the deblurring effect, and there are other problems such as poor stability of these algorithms, which cannot overcome the lack of ambiguity. Targeting at these problems, we proposed a Bayesian model coupled non-stationary priors image deblurring mechanism. Methods Based on second-order statistics, vague function was defined. Filtering factor and ultra-parameter structure were introduced to maintain a priors model of non-stationary edge preserving. Based on Bayesian inference, Jacobian matrix was introduced to design the hyperparameter dynamic update mechanism. The coupled priors model and Bayesian model were used to complete image reconstruction. The performance of the algorithm was tested on the simulation platform. Results Compared with several other mechanisms, the mechanism proposed in this paper showed better deblurring performance. The texture details remained clear after local amplification, and the structural similarity of the images before and after deblurring was higher. Conclusion The proposed algorithm had relatively good image deblurring performance and the reconstruction quality was ideal.