to appear soon
Description PETMEI 2016:
Let I[r; c] be a digital close-up image of the eye in the nearinfrared
spectrum with r rows and c columns. The eyelid
detection task consists of selecting two sets of pixels Pl and
Pu in I that lie respectively on the lower and upper eyelids,
which are then used to fit functions representing the outline
of each eyelid.
The proposed method consists mainly of I) rescaling the image
preserving dark regions to reduce noise and computation
costs, II) filtering the image according to a combination
of local features to generate a likelihood map for the eyelids,
III) detecting edges on the likelihood map, and selecting
two edges to represent the eyelids based on their orientation
and horizontal shift in respect to one another, enclosed
intensity value, and accumulated likelihood. These
steps are described in detail in the following subsections,
followed by a graphical representation exemplifying the output
of each stage in the algorithm.
Results eyelid detection:
Outline similarity with the jaccard index(higher is better).
Eyelid aperture estimation (lower is better).
Cumulative detection rate (top and left is better).
WACV 2017 (on permission)
All algorithms (ftp)
- Eyelid detection WACV 2017
- Eyelid detection PETMEI 2016
Data set downloads:
All data sets (ftp)
- Eyelid detection data set WACV 2017
- Eyelid detection data set PETMEI 2016