Image deblurring is of great importance for wide applications in reality. An existing method based on the dark channel prior (DCP) has achieved notable success across natural, facial, and text images. However, this method encounters challenges when applied to images with large kernels or more saturated pixels. The main reason of this limitation may come from the linear degradation assumption, while the blur process is often more complex and against the linear assumption. This work introduced an intermediate image correction strategy to enhance the DCP-based deblurring method. By estimating the probability of intermediate image pixels deviating from the linear degradation model, we selectively corrected the disadvantageous pixels. In this way, we can effectively reduce the unfavorable structures and thus improve the accuracy of blur kernel estimation. Extensive experiments show that the intermediate correction strategy enables the proposed model to restore sharp image details more effectively, resulting in significant improvement for blind deblurring performance.