Quality assessment of eye fundus images taken by wide-view non-mydriatic cameras

Non-Mydriatic fundus cameras are very suitable for teleophthalmology framework, because this type of fundus cameras does not require eye-drop to dilate patient's pupil to take image, and then image acquisition can be realized without expert such as ophthalmologist. However, in this scheme, automatic and accurate image quality assessment on site is indispensable, because low-quality images are useless for reliable diagnostic. In this paper we analyze several generic features, such as statistic feature of histogram, cooccurrence matrix, run-length and Cumulative Probability of Blur Detection (CPBD), for automatic quality assessment of the fundus images taken by wide-view non-mydriatic fundus cameras. The performance of several combination of the generic features extracted from the fundus images are evaluated using different classifiers, such as support vector machine, K Nearest Neighbor (KNN) classifier, decision tree-based classifiers, etc.