Eight eigenfaces
illustration_link (57 Kbytes)
Photobook creates a reference set of facial characteristics by analyzing all the faces in a database. Rather than trying to match features humans might remember (e.g., hair color or the shape of a nose), Photobook uses essential characteristics of each facial image, called eigenfaces, to provide the patterns with which to compare a target face. The images above show the first eight eigenfaces in one reference set.
Composite of the average face
illus
tration_link (21 Kbytes)
The program builds a composite of a face using all the eigenfaces in a database (in this case, 7000 images) to represent the average face.
Facial-recognition search
illustration_link (35 Kbytes)
The screen above appears at the end of a Photobook facial-recognition search to present the closest matches (in descending order) to the target face (encircled in red in the upper left corner). Note that the target and the top three matches are different images of the same individual.
Uncompressed
illustration_link (30 Kbytes)
In addition to accuracy, Photobook's eigenface model is also an efficient way to store images. The eigenface image here requires only 85 bytes of storage space
JPEG-Compressed
illustration_link (26 Kbytes)
The murkier image here is a JPEG-compressed version of the original that requires 540 bytes of storage.
How Photobook Recognizes Faces
