Blink Identification Using Video Magnification for Temporal Analysis of Eyelid Motion

This work proposes a blink detection method based on video magnification and evaluates its performance against the Eye Aspect Ratio (EAR), a widely used geometric landmark based approach. The methods were assessed using 221 manually annotated blink events collected from two subjects performing fixation and reading tasks at three acquisition distances. The proposed method achieved an overall F1-score of 0.997, with 100% recall in nearly all scenarios, whereas EAR reached an F1-score of 0.971 and showed systematic failures in detecting partial blinks and estimating event duration. The results indicate that representing blinking as a continuous temporal signal improves detection performance while preserving morphological information relevant to subtle blink patterns.