A Machine Learning Approach for Identifying Five Types of Horizontal Ocular Disorders Using Haar Features

Abstract

Background: The proliferation of digital devices, such as smartphones, TVs, tablets, and laptops, has raised concerns about the potential long-term impact on eye health, particularly from blue light emitted by screens. Related studies suggested that prolonged exposure to blue light may contribute to visual impairments or discomfort. Objective: This research introduces an innovative machine learning approach aimed at diagnosing such visual impairments by automatically detecting the iris center in images using a combination of the Haar Cascade Classifier and Circular Hough Transform algorithm. Methods: The proposed methodology has three phases. The first is the development of a diverse facial image dataset from various datasets through three steps: designing an AdaBoost-based cascaded classifier that identifies the face and the region around the eyes; subsequently, iris boundaries are labeled, where an optimized Hough Transform approach determines its center. Further, the model segments the eye into three regions to ascertain the iris position cause for the deviation of eyes. Finally, RBFN is used to classify five classes of eye strabismus with high accuracy on a horizontal axis. Results: The proposed system presents very promising results with an accuracy equal to 97.5%, precision 0.97834, specificity 0.33333, and recall 0.99632 in iris center coordinates extraction of five classes of horizontal strabismus classification. These findings were validated over two different datasets in order to prove the strength of this model in detecting eye disorders under different light and photo scenarios. Conclusions: This method may be acting like a bridge between computer vision and ophthalmology in ocular health assessment and treatment. Further improvements may open further horizons regarding automated eye detection and diagnosis.