Restoring the true image from a degenerated version is a classical ill-posed inverse problem. Regularization technique is one of the mainstream methods to solve this problem. It restricts the solution image into a regularization space, and the restored image is the projection of the degenerated image in this space. However, it is difficult to choose a proper regularization space for different images. In order to improve the adaptability of the regularization model and model images with different features more precisely, an adaptive hybrid regularization model based on fuzzy set theory is proposed in this paper. Firstly, learning algorithm is used to calculate the membership degree of the images in different regularization spaces. And then the first s spaces with the largest membership degree are selected to establish an adaptive hybrid regularization model with membership degree as the weights. Finally, ADMM (Alternating Direction Method of Multipliers) based algorithm is used to numerically solve the model. The experimental results validate the proposed model. For different synthetic and real images, the proposed model can effectively choose the appropriate regularization spaces and obtain satisfactory results.