March 2012
Volume 53, Issue 14
Free
ARVO Annual Meeting Abstract  |   March 2012
A Preprocessing Framework for Low-Cost Retinal Fundus Images
Author Affiliations & Notes
  • Thomas Köhler
    Pattern Recognition Lab,
    Graduate School in Advanced Optical Technologies (SAOT),
    University of Erlangen-Nuremberg, Erlangen, Germany
  • Bernhard Höher
    Graduate School in Advanced Optical Technologies (SAOT),
    Chair for Microwave Engineering and High Frequency Technology,
    University of Erlangen-Nuremberg, Erlangen, Germany
  • Joachim Hornegger
    Pattern Recognition Lab,
    Graduate School in Advanced Optical Technologies (SAOT),
    University of Erlangen-Nuremberg, Erlangen, Germany
  • Georg Michelson
    Graduate School in Advanced Optical Technologies (SAOT),
    Department of Ophthalmology,
    University of Erlangen-Nuremberg, Erlangen, Germany
  • Footnotes
    Commercial Relationships  Thomas Köhler, None; Bernhard Höher, None; Joachim Hornegger, None; Georg Michelson, None
  • Footnotes
    Support  Erlangen Graduate School in Advanced Optical Technologies (SAOT), Medical Valley EMN Leading-Edge Cluster
Investigative Ophthalmology & Visual Science March 2012, Vol.53, 4061. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Thomas Köhler, Bernhard Höher, Joachim Hornegger, Georg Michelson; A Preprocessing Framework for Low-Cost Retinal Fundus Images. Invest. Ophthalmol. Vis. Sci. 2012;53(14):4061.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract
 
Purpose:
 

Our goal is to provide a computational framework for image quality improvement of low-cost fundus images to make them usable for medical applications.

 
Methods:
 

We use an industrial CCD camera to capture uncompressed sequences of fundus images with a frame rate of 2fps. The images cover the eye background with a resolution of 1280x960 pixels. Each color image is acquired by a sequential acquisition of three frames representing the RGB color channels via illumination by three different LEDs. This method increases the photosensitivity of the system while the light exposure of the human eye is decreased and the brightness of each color channel can be controlled independently.For each frame illumination correction is applied. This creates homogenous illumination and equalizes reflections on the eye background. The focus of our framework is denoising and super-resolution to reduce noise and to increase the spatial resolution. We propose a temporal denoising scheme which is based on adaptive frame averaging. Since the human eye moves during the acquisition in a stochastic manner image registration is performed to align all frames. The motion compensation is done by mutual information based rigid registration. Preliminary results were obtained with three image sequences of 18 frames for each RGB color channel (see figure (a) for the green channel).

 
Results:
 

Illumination correction homogenizes the illumination (figure (b)). Temporal denoising increases the signal-to-noise ratio with respect to the first frame by 145.5%, 166.8% and 139.8% for the red, green and blue channel respectively. Structural details such as blood vessels are preserved (figure (c)). The RGB-composite contains color information of three grey-scaled frames (figure (d)).

 
Conclusions:
 

The proposed framework enables a significant quality improvement of low-cost fundus images. Illumination correction and denoising are initial steps towards practical applications of a low-cost fundus camera. In our future work we will extend denoising to super-resolution to increase the spatial resolution.  

 
Keywords: image processing • imaging/image analysis: non-clinical 
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×