| Title |
Classification of Cataract using Frequency Domain Features and Deep Learning Method |
| Authors |
신민재(Minjae Shin) ; 김대원(Daewon Kim) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.2.67 |
| Keywords |
Cataract; Fundus image; Frequency domain; Convolutional neural network; Deep learning |
| Abstract |
A disease in which the eyes become cloudy and difficult to see is called a cataract. A doctor's judgment is required to diagnose the cataract but the criteria for diagnosing the cataract may differ for each doctor so the cataract diagnosis result may be different. Therefore, it is necessary to diagnose the cataract according to consistent criteria. In this paper, we conducted a study to diagnose the cataract using an artificial intelligent computing algorithm. The fundus image of a normal eye appears clear but the image of a cataract eye looks slightly blurred. This means that there is a difference in frequency components for each image. Therefore, in this study, using these features, we converted the fundus images into frequency domain images and used them as input data. Next, we classified the cataract using a Convolutional Neural Network (CNN) which is an artificial intelligence technique used for image classification. The CNN shows differences in performance depending on the hierarchical structures. We designed several CNN algorithms of various structures, measured their performances, and selected the model with the best results and presented it as the final result. This study includes the process of Fourier Transform and Discrete Cosine Transform (DCT) of the fundus images, and the final study result showed a classification accuracy of 95.83%. |