In the context of self-medication, we propose a novel, smartphone-based method for accurately detecting symptoms of dry eye, an increasingly common and often underdiagnosed condition. This approach utilizes a specially designed device that combines a smartphone with an adjustable ring light to capture high-quality images of the eye, focusing on the tear meniscus region. By analyzing the height and shape of the crescent-shaped reflection within the tear meniscus, we estimate tear volume to assess potential dry eye symptoms and severity. Additionally, computational adjustments are applied to correct errors caused by variations in shooting distance, enhancing detection reliability and consistency. To further improve detection accuracy, we introduce an innovative binarization technique using the U-Net deep learning model, enabling precise, localized binarization specifically within the crescent reflection area. Compared to traditional methods, such as Otsu’s and adaptive binarization, this technique achieves significantly higher accuracy, nearing the precision typically associated with medical-grade imaging equipment.