No abstract

No abstract

No abstract

<p>In this work we develop a compact neural network that is designed to deblur images that have been affected by a non-uniform blur. We develop this network by unrolling a traditional iterative image deblurring algorithm and adapt it to independently deblur regions of an image. The network is evaluated by comparing its deblurring capabilities with that of other state-of-the-art networks, such as the SRN-Deblur network [13] and the DUBLID network [4]. We investigate the effect that varying the patch size and the size of the point spread function has on the deblurring performance of our network. We evaluate each deblurring network using the industry standard GOPRO [12] and Kohler [25] datasets.</p>

<p>In this work we develop a compact neural network that is designed to deblur images that have been affected by a non-uniform blur. We develop this network by unrolling a traditional iterative image deblurring algorithm and adapt it to independently deblur regions of an image. The network is evaluated by comparing its deblurring capabilities with that of other state-of-the-art networks, such as the SRN-Deblur network [13] and the DUBLID network [4]. We investigate the effect that varying the patch size and the size of the point spread function has on the deblurring performance of our network. We evaluate each deblurring network using the industry standard GOPRO [12] and Kohler [25] datasets.</p>