Person identification based on low quality eye images using machine learning for cardiac procedures

Abstract

Eye recognition for person identification has been studied and implemented in many healthcare settings. This contactless person identification system is useful, especially in a hospital during the COVID-19 pandemic. It is also useful in critical treatment management, such as cardiac procedures. Unfortunately, low quality eye images may be obtained, which can result from, for example, a low-resolution camera on a mobile phone or moving eyes. In this study, a method to identify person in hospitals based on low-quality images is proposed. The eye images are captured using a low-resolution camera or moving camera. The proposed system employs an image segmentation algorithm and compares three different machine learning approaches to effectively classify each segmented region as the appropriate recognition type using Cascade Trainer: neural networks, support vector machines, and random forest decision trees. The use of a wrapper technique combined with recursive feature reduction has proven to be successful in maintaining the classifiers' performance while considerably lowering the number of required predictors. The results obtained with Jupyter demonstrate that classifiers based on fitted neural networks, random forest models, and support vector machines achieve high overall accuracy on testing with significant differences. The purpose of this project is to provide a consistent and robust methodological framework for the creation of trustworthy computational systems to assist in eye recognition for patient identification by using standard classification methods. The preliminary findings of this study suggest that, based on eye images of the classes collected in the dataset, they can be autonomously recognize.