Correcting Automatic Cataract Diagnosis Systems Against Noisy/Blur Environment

Abstract

In this paper, a methodology to improve the performance of existing automatic cataract detection systems (ACDS) in noisy/blur environment is proposed. The presented approach consists of dual-threshold based image quality evaluation module to enhance the performance diminution of ACDS in noisy/blur environment. Initially the first threshold is obtained from naturalness image quality evaluator (NIQE) and then second threshold is achieved through noise level estimation (NLE). In order to ensure robustness, the proposed method is evaluated with artificially created noise and blur datasets in association with existing pre-trained convolution neural network based ACDS. The experiments results show superiority in performance over existing methods in literature.