Retinal image quality assessment using generic image quality indicators

A retinal image gradability assessment algorithm based on the fusion of generic image quality indicators is introduced. Four features quantifying image colour, focus, contrast and illumination are computed using novel image processing techniques. These quality indicators are also combined and classified to evaluate the image suitability for diagnostic purposes. The algorithm performance is thoroughly appraised through comparison of the automatic classification results of 2032 retinal images from proprietary, DRIVE, Messidor, ROC and STARE datasets with human made classification, revealing a sensitivity of 99.76% and a specificity of 99.49%. The algorithm computational complexity and sensitivity to image noise and resolution were also experimentally quantified demonstrating very good performance and confirming the usability of the solution in an ambulatory application environment. (C) 2012 Elsevier B.V. All rights reserved.