Eyelign : A Holistic Framework for Automation of Strabismus Detection, Classification, and Accurate Positional Analysis.

Abstract

"The primary issue is on the vital requirement for a precise, effective, and easily available diagnosis of strabismus, a condition marked by ocular misalignment that affects vision, depth perception, and may result in social stigma. Conventional diagnostic techniques, which include manual exams by experts, have drawbacks including subjectivity, patient pain, time commitment, and restricted accessibility, especially when it comes to limited resources or remote locations. This issue emphasizes how important it is to have an automated, accurate, and user-friendly system that can recognize and categorize digital face photos for strabismus. The creation of a method to accurately diagnose strabismus is of utmost importance in improving patient care and ensuring accessibility to necessary diagnostic services. This is because determining the type and degree of strabismus is crucial for formulating successful treatment options. This research project uses the dlib library's facial landmark detection to present a novel technique for improving the iris color in facial photographs. The method allows for accurate segmentation of the eyes by creating exact eye masks based on landmarks that have been recognized. In order to create more precise iris masks, centroid determination is performed after thresholding on the red channel for iris recognition. After that, a color modification method is used to make the iris region seem better. When combined with the original image, the result is visually pleasing. Through class and sequence diagrams, the research shows a thorough architecture design while also discussing system needs and possible societal ramifications. Added to the changing field of face image processing technology is a comparative examination of advantages and disadvantages. The study was able to attain an accuracy of 89.39% and 95.83% for pupil detection with the eyelign dataset and BoiID face dataset respectively, along with F1 scores of 0.9586 and 0.9881. For the strabismus detection module, the study achieved an accuracy of 89.43% while strabismus classification only heightened to an accuracy of 74.797%."