Ocular multi-spectral imaging deblurring via regularization of mutual information

Abstract Multi-Spectral Imaging is one non-invasive technique recently introduced in ocular disease diagnosis. Its performance significantly correlates with the imaging quality. We observed that various degenerations including motion and out-of-focus blurry effects had degraded the quality of ocular MSI images. Bearing this in mind, we propose a multi-modality, multi-image deblurring framework through integrating the information from the aligned images since the accurate correspondence between each sequence of the ocular images need to be constructed simultaneously. The smoothness of ocular MSI image and its corresponding blur kernel are simultaneously taken as regularization terms. Meanwhile, to leverage the complementary information within a set of ocular MSI images, the mutual information between each pair of ocular MSI images is also exploited as a vital regularization term in the presented optimization framework. To evaluate the performance of the proposed approach, we conducted comparison experiments between the state-of-the-art techniques and ours. Experimental results show that the proposed technique outperforms state-of-the-art deblurring methods quantitatively and visually.