Abstract
Sophisticated mathematical algorithms (such as differencing, thresholding, aggregation and statistical analysis of skin colours) are used to compare successive frames of computer-captured images of the face. From these, changes in state of the eyes are determined and are used to detect blinks. A recognition performance of 83.74 +/- 0.03% is achieved over five subjects with a low rate of false positives 2.71 +/- 0.01%. A logical decision rule identifies purposeful blinks and applies them to control either a custom-designed communication package or an external device.