Contrast measurement for no-reference retinal image quality assessment

Abstract

Retinal fundus image provides information of retinal pathologies to diagnose some diseases by computer automatic detection. The quality of the retinal image mostly affects the detection results. In this research, blood vessels contrast measurement algorithm is approached as the first step in no-reference retinal image quality metric. The step includes segmentation of blood vessels. This work was used retinal images from HEI-MED database. The retinal images are divided as poor and good quality, and then compared with the expert assessment. The result shows that the performance of the approach algorithm is correlated closely with the expert assessment. The qualitative evaluation achieves sensitivity 0.97619, specificity 0.8 and accuracy 0.89362.