A novel framework method for non-blind deconvolution using subspace images priors

Non-blind deconvolution has been an active challenge in the research fields of computer vision and computational photography. However, most existing deblurring methods conduct direct deconvolution only on the degraded image and are sensitive to noise. To enhance the performance of non-blind deconvolution, we propose a novel framework method by exploiting different sparse priors of subspace images. In the proposed framework, three effective filters are firstly designed to decompose a degraded image into the measurements of different subspace images. Then, existing deblurring techniques are employed to deblur different blurred subspace images respectively. Finally, the least square integration method is utilized to recover the ideal image by integrating the deblurred estimates of subspace images with the degraded image. The proposed framework is more general and can be easily extended to existing deblurring methods. The conducted experiments have validated the effectiveness of the proposed framework, and have demonstrated that the proposed method outperforms other state-of-the-art methods in both preserving image structures and suppressing noise.