09-08-2012, 03:47 PM
Pose Discriminiation and Eye Detection Using Support Vector Machines (SVM)
Pose Discriminiation and Eye Detection Using.pdf (Size: 187.92 KB / Downloads: 29)
Abstract.
Most face recognition systems assume that the geometry of the image
formation process is frontal. If additional poses, beyond the frontal one, are
possible, then it becomes necessary to estimate the actual imaging pose. Once a
face is detected and its pose is estimated one proceeds by normalizing the face
images to account for geometrical and illumination changes, possibly using
information about the location and appearance of facial landmarks such as the
eyes. This paper describes a novel approach for the problem of pose estimation
and eye detection using Support Vector Machines (SVM). Experimental results
using frontal, and 33.75o rotated left and right poses, respectively, demonstrate
the feasibility of our approach for pose estimation. The image (face) data comes
from the standard FERET data base, the training set consists of 150 images
equally distributed among frontal, 33.75o rotated left and right poses,
respectively, and the test set consists of 450 images again equally distributed
among the three different types of poses. The accuracy observed on test data,
using both polynomials of degree 3 and Radial Basis Functions (RBFs) as kernel
approximation functions, to determine the SVM separating hyperplanes, has been
100%. On the eye detection task, the training data consisted of 186 eye images
and 186 non-eye images. SVM was tested against 200 test examples (eye and
non-eye). The best generalization performance of 4% was achieved using
polynomial kernels of second degree as the set of approximating functions. SVM
appear to be robust classification schemes and this suggests their use for
additional face recognition tasks, such as surveillance.
Key Words: eye detection, face recognition, FERET, pose estimation, predictive
learning, Radial Basis Functions (RBFs), Support Vector Machines (SVMs).
Proceeding of NATO-ASI on Face Recognition: From Theory to Applications, 1998
1 Introduction
There are several related (face recognition) sub problems: (i) detection of a
pattern as a face (in the crowd) and its pose, (ii) detection of facial landmarks,
(iii) face recognition - identification and/or verification, and (iv) analysis of facial
expressions (Samal et al., 1992). Face recognition starts with the detection of
face patterns in sometimes cluttered scenes, proceeds by normalizing the face
images to account for geometrical and illumination changes, possibly using
information about the location and appearance of facial landmarks, identifies the
faces using appropriate classification algorithms, and post processes the results
using model-based schemes and logistic feedback (Chellappa et al, 1995). Most
face recognition systems assume that the geometry of the image formation
process is frontal. If additional poses, beyond the frontal one, are possible, then it
becomes necessary to estimate the actual imaging pose. Pose information can
then be used in a variety of ways, ranging from specific normalization and facial
landmark detection procedures, to face recognizers trained for some specific pose
only. This paper addresses the problems of pose estimation and eye detection
using Support Vector Machines (SVM), a novel pattern classification algorithm
(Cortes and Vapnik, 1995; Vapnik, 1995).
2 Predictive Learning
The goal of predictive learning is to develop a computational relationship for
estimating the values for the output variables given only the values of the input
variables. Different taxonomies are available for predictive learning, among them
that including regression and density estimation, and classification, corresponding
to the input variables being continuous and discrete / categorical, respectively,
even that any classification problem can be reduced to a regression problem. Pose
detection is a classification problem and this paper describes novel and robust
means for solving it. Predictive learning can be thought of as a relationship y =
f(x) + error, where the error is due both to (measurement) noise and possibly to
’unobserved’ input variables. The main issues one has to address are related to
prediction (generalization) ability, data and dimensionality reduction
(complexity), explanation / interpretation capability, and possibly to biological
plausibility.
In the framework of predictive learning, estimating (’learning’) a model
(’classifier’) from finite data requires specification of three concepts: a set of
approximating functions (i.e., a class of models : dictionary), an inductive
principle and an optimization (parameter estimation) procedure. The notion of
inductive principle is fundamental to all learning methods. Essentially, an
inductive principle provides a general prescription for what to do with the
training data in order to obtain (learn) the model. In contrast, a learning method
is a constructive implementation of an inductive principle (i.e., an optimization or
parameter estimation procedure) for a given set of approximating functions in
which the model is sought (such as feed forward nets with sigmoid units, radial
basis function networks etc.) (Cherkassky and Mulier, 1998). There is just a
handful of known inductive principles (Regularization, Structural Risk
Minimization (SRM), Bayesian Inference, Minimum Description Length), but
there are infinitely many learning methods based on these principles. It is the
prediction risk, the expected performance of an estimator (’classifier’) for new
(future) samples, which determines to what degree adaptation has been successful
so far.
Accurate estimation of prediction risk from available training data is crucial
for the control of model complexity (model selection). Classical methods for
model selection are usually based on asymptotic results for linear models. Nonasymptotic
(guaranteed) bounds on the prediction risk based on VC-theory have
been proposed by Vapnik (1995) as part of Statistical Learning Theory (SLT).
Prediction risk, for regression problems in general, and classification problems in
particular, is usually estimated as a function of the empirical risk (training error)
penalized (adjusted) by some measure of model complexity (see also the known
tradeoffs between bias and variance). Once an accurate estimate of the prediction
risk is found it can be used for model selection by choosing the model complexity
which minimizes the estimated prediction risk. The Support Vector Machines
(SVM), discussed in the next section, build pattern classifiers using SRM as the
inductive principle, Radial Basis Functions (RBFs) or polynomial splines (as
possible sets of approximating functions), and dual quadratic optimization. The
constructive implementation of the SRM principle used by SVM is to keep the
value of the empirical risk fixed (small) and minimize the confidence interval of
the predicted risk.