List of Figures

• 3D Normalization as a Bridge Between Feature Detection and Face Recognition
• Sobel edge detection
• Deriche edge detection at multiple scales
• Typical output from Sobel edge detection
• Post-Processing Sobel edge detection
• An input image pyramid for multi-scale edge extraction
• Multi-Scalar edge extraction with the Sobel operator
• Cocircularity of edges
• The set of circular sampling regions, the set of symmetry orientations and the set of cocircular edge orientations
• Discarding edges misaligned with the annulus normals
• Input to the attentional mechanism
• Lines of general symmetry at multiple scales
• Combining lines of symmetry
• The semi-elliptical model
• A sample template for face or head-like blob detection with template normals represented with intensity
• Projecting onto template normals to attenuate misaligned edges
• Splitting templates into angular bins
• Computing the angular profile of a contour
• Symmetric enclosure versus traditional template matching
• The hierarchical search sequence for faces and facial features
• The multi-scalar interest map pyramid
• The search space for the selective symmetry detector
• Insufficient operator overlap problems
• The face templates used by the selective symmetry detector
• The collection of detected possible facial contours
• Generating eye search region from the facial contour
• Isolating the eye search regions
• Eye operator size versus face size
• Isolating the eye search regions
• Strong symmetry responses from structures other than eyes
• Eye midpoint not centered in facial contour
• Minimum intra-ocular distance
• Horizontal Projection of Symmetry Points
• Limb extraction
• Excessive limb undulation
• Search space for seeding mouth limb extraction
• Search space for mouth superimposed on face
• Limb axes extracted as 3D trajectories
• Intensity variance
• Intensity variance with mouth locus
• The segmented mouth limb
• Nose edge data
• Nose Height
• Finding the nose tip from the nose bottom using the edge signature.
• The nose-bottom-line and the nose-tip-line.
• Iris Localization
• A typical output of the face detection stage.
• In-plane rotation, scaling and translation.
• Out-of-plane or depth rotations.
• Some of the 3D faces used to form the average 3D face.
• The average 3D face
• Image of U.S. President Ford with eyes, nose and mouth located
• The scaled orthographic projection of the model upon the image plane
• Stretching the model in search of the best mouth fit.
• The 3D model coated with an intensity image's face.
• Mirroring intensity images from one side of the face to the other.
• Range of nose positions where mirroring is necessary.
• Some re-projections of the coated 3D model.
• A synthesize mug-shot image of U.S. President Ford.
• Histogram of the mean face.
• Windowing to split face histogram correction for each side of the face.
• Limiting histogram generation to avoid hair and beards.
• The mixture of histogram mappings from the left to the right side of the face.
• The mean face.
• The ordered eigenvalues of the dataset.
• The ordered eigenvectors (or eigenfaces) of the dataset.
• The distribution of the dataset in the first three coefficient dimensions.
• The mean face generated with smaller mug-shots.
• The distribution of first two coefficients and the residue (on the vertical axis) for the dataset.
• Mug-shots containing true faces and non-faces and a graph of their distance to face space (DFFS) values.
• The nose localization problem.
• Distance to face space values for mug-shot images from different nose-point trials (from left to right across the nose-line)
• The final localization.
• Block diagram description of the overall algorithm
• The video output for a sample image
• The video output for a sample camera snapshot
• The 30 individuals in the Achermann database
• The 10 different views per individual in the database
• The recognition video output
• The recognition rates for different views
• The recognition rates for different individuals
• Synthesized mug-shots with left eye anchor point perturbed
• KL approximations to mug-shots with left eye anchor point perturbed
• Variation in recognition certainty under left eye anchor point perturbation
• Variation in recognition certainty under right eye anchor point perturbation
• Variation in recognition certainty under nose anchor point perturbation (View 1)
• Variation in recognition certainty under nose anchor point perturbation (View 2)
• Variation in recognition certainty under mouth anchor point perturbation