Peer Reviewed

1

DOI

10.17077/omia.1036

Conference Location

Munich, Germany

Publication Date

October 2015

Abstract

Fundus retinal imaging is widely used in the diagnosis and management of eye disease. Blur commonly occurs in the acquisition and when it is severe the resulting loss of resolution hampers accurate clinical assessment. In this paper, we present a new technique to address this challenging problem. We make use of implicitly constrained image deblurring, which is known to provide improved results over unconstrained and explicitly constrained methods, and build this into a multi-channel variational framework for parametric deblurring. We propose a new method for automatically selecting the regularisation parameter in the absence of the true (sharp) image using vessel segmentation. We then modify the model to include a regularisation coefficient function which is dependent on an available image mask in order to avoid potential inaccuracies caused by the addition of artificial masks. We present experimental results to demonstrate the effectiveness of our new method.

Rights

Copyright © 2015 Bryan M. Williams, Ke Chen, Simon P. Harding, and Yalin Zheng

Included in

Ophthalmology Commons

Share

COinS
 
Oct 9th, 12:00 AM Oct 9th, 12:00 AM

A New Method of Blind Deconvolution for Colour Fundus Retinal Images

Munich, Germany

Fundus retinal imaging is widely used in the diagnosis and management of eye disease. Blur commonly occurs in the acquisition and when it is severe the resulting loss of resolution hampers accurate clinical assessment. In this paper, we present a new technique to address this challenging problem. We make use of implicitly constrained image deblurring, which is known to provide improved results over unconstrained and explicitly constrained methods, and build this into a multi-channel variational framework for parametric deblurring. We propose a new method for automatically selecting the regularisation parameter in the absence of the true (sharp) image using vessel segmentation. We then modify the model to include a regularisation coefficient function which is dependent on an available image mask in order to avoid potential inaccuracies caused by the addition of artificial masks. We present experimental results to demonstrate the effectiveness of our new method.