View abstract

Session S20 - Applied Math and Computational Methods and Analysis across the Americas

Friday, July 16, 12:00 ~ 12:30 UTC-3

On the implementation of mimetic anisotropic filtering with convolutional neural networks for glaucoma suspect detection

Lola Bautista

Universidad Industrial de Santander, Colombia   -   This email address is being protected from spambots. You need JavaScript enabled to view it.

Anisotropic filtering has been widely studied for edge detection in digital image processing. Special attention has been given to the Perona-Malik model in multiscale image analysis for image restoration. In a recent work [1] it was developed an implementation of such filtering through mimetic methods which discretise the gradient and divergence operators of the diffusion coefficient in the Perona-Malik model, which showed better effectiveness of glaucoma suspects detection when the filter was applied to the input images. In the recent years, convolutional neural networks (CNN) had become an important tool in image processing because their capabilities to learn the internal representation of images. In this work we present the comparison of two pre-trained CNNs using the GoogleNet architecture with a training set of 1084 images each, collected by the Centro de Prevención y Consultoría en Glaucoma, in order to estimate the influence of the mimetic anisotropic filtering before training. The first CNN was trained without the filter, and the second one was trained with the images after applying the mimetic anisotropic filter, reaching an accuracy of 90.68\%. It remains open the problem of calibrating the parameters of the filter, as well as the stopping criterion to avoid degradation of the image.

References [1] Jorge Villamizar et al. “Mimetic Finite Dierence Methods for Restoration of Fundus Images for Automatic Detection of Glaucoma Suspects.” In: Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization (2021). doi:10.1080/21681163.2021.1914733.

Joint work with Jorge Villamizar (Universidad Industrial de Santander, Colombia; Universidad de Los Andes, Venezuela, Biomedical Imaging, Vision and Learning Laboratory, Colombia), Juan Carrillo (Biomedical Imaging, Vision and Learning Laboratory, Colombia) and Juan Rueda (Centro de Prevención y Consultoría en Glaucoma, Colombia).

View abstract PDF