Fig. 3
From: Identification of ocular refraction based on deep learning algorithm as a novel retinoscopy method
Diagram of the system construction. A Image pre-processing and augmentation; a original RFPs; b the largest inscribed circle and rotation; c rotation and sharpness; d contrast-limited adaptive histogram equalization was used to improve colour and spatial contrast between the structures and the background retina for RFPs; e histogram equalization processing; f the largest inscribed square. B Confusion matrix between target values and eigenvectors. The colours in the figure indicate the strength of the correlation. C Architecture of the FMDLS proposed in this study. (Upper picture) the pipeline for sphere and cylinder, a and b were networks for two different models, c was the classification model; (Bottom picture) the pipeline for axis, a and b were two different classification models