Accelerating Blind Deconvolution: A Preprocessing Approach for Point Spread Function Size Estimation

Existing approaches in blind image deblurring often take the blur kernel size as a manual parameter, and when this parameter is not set correctly, it can lead to significant errors in the estimated kernel. Given that multiple other parameters also influence each deblurring operation with a specified kernel size, manually determining the kernel size is a tedious task and often results in trial and error for the user to find the best size. Therefore, proposing an automatic method for the initial estimation of the kernel size can be beneficial as a preprocessing step in blind deblurring operations. In this paper, we propose a new approach for the automatic estimation of the kernel size. In this approach, we utilize the quality curve of deblurred images with kernels of different sizes to find the optimal point. In this process, a blind deblurring method is employed to deblur the image, and a no-reference quality assessment metric is used to evaluate the quality of the deblurred images. The results show that the proposed method can accurately estimate a suitable kernel size.