A multi-local kernels′ fusion algorithm is proposed to solve the problem of high complexity of dark channel priors. In this algorithm, the local kernels are computed and solved in parallel, and then merged into a global kernel (point spread function) by using the shape similarity of local kernels. For the noise that has appeared on the initially merged global kernel, the relevance adjustment is introduced using the neighboring information to further improve the fusion effect. Experiments show that the proposed algorithm can effectively improve the speed of image deblurring while guaranteeing the deblurring effect. It also has better effect on the local detail restoration of some real blurred images, and can handle large-size blurred images well.