Pupil Positioning via Semi-Supervised Learning

The patent combines unsupervised learning and supervised learning into a semi-supervised learning framework. The unsupervised learning component processes the input image to generate an initial pupil region mask, while the supervised learning component refines this mask using labeled training data. This merging allows the system to maintain processing simplicity from unsupervised learning while achieving high positioning accuracy through supervised learning refinement.

The unsupervised learning process performs preliminary action by generating an initial pupil region mask before the supervised learning refinement. This preliminary mask provides a good starting point that reduces the search space for the supervised learning component, making the overall process more efficient while maintaining high accuracy.

The patent segments the processing into two distinct stages: unsupervised learning for initial mask generation and supervised learning for refinement. This segmentation allows each component to be optimized independently - the unsupervised learning handles the complex initial processing while the supervised learning focuses only on refinement, reducing overall complexity compared to using supervised learning for the entire process.

The supervised learning component performs only partial action by refining the pupil mask rather than generating it from scratch. This partial refinement approach reduces the complexity burden on the supervised learning component while still achieving high positioning accuracy, as the heavy lifting is done by the unsupervised learning stage.