Loading ...
Aged face detection and recognition using splines
Aug 2nd, 2007 by admin
Aged face detection and recognition using splines
Abstract
Face recognition has been one of important subject
of interest owing to its applications in various
streams of technology. May it be a security purpose
or data mining with pattern recognition, every work
demands this issue. It can serve for imitation or
photographic applications, or recognizing disguised
faces.In this paper we intend to represent the
mechanism of generating faces with age as the factor.
In case any photograph is provided, the basic
features can be mapped and estimated in terms of
curve sections or graphics primitives and thereby
certain measures be taken care to generate aged
faces. Also in accordance to basic features, the face
can be generated with the help of smooth curves
which best defines the face. Thus the basic
mechanism of recognition works with detection with
growing age. There may be several instances when
the detection phenomenon may need an estimated
image for about 10 years from now. In such practical
cases the properties of the curves are such
manipulated so as to generate the past or future
impression.
(Note: This Paper was presented in ICSIP., Signalspot.
Please Download the paper for proper formatting, images and symbols)
Keywords: Splines, Snakes, active contour
models, matching and recognition, Eigen vector,
Introduction
Conceptually the idea of face recognition is denoted
in terms of face identification and verification. It
basically means identifying any individual against a
number of other people. There have been several
works based on recognizing objects which are partly
hidden from view or occur against cluttered
backgrounds[1]. This is accomplished by comparing
certain features of images like intensity edges which
eliminates the need of comparing the whole of
images. The method was popularly worked with
Eigenspace View Matching techniques. Similarly the
face recognition system is also analyzed in terms of
face localization and segmentation, normalization
and finally actual face identification or verification
(also termed as authentication). Hereby the features
are extracted and classified. [2]
Past researches have dealt with the identification procedures
based on geometrical features like eyebrow thickness, width of
nose, set of radii describing the shape of chin, face width at
nose position etc. Another system of identification is to
normalize the faces to be matched, sub images of the face
need to specify in terms of facial features like nose, eyebrow,
eyes, and many others.
Our concept lies in generating faces of the same person with
ageing so that the face is recognized irrespective of time
elapsed. In most of the cases where face recognition is to be
done, the factor that arises is the change that corresponds with
growing age. A face in young age is smooth in case
graphically analyzed, while it changes in appearance as the
time passes. Also in case of disguised appearance, the face
may adopt changes in texture or any other feature to avoid
identification. What remains unchanged is the basic structure
of skull and face that defines on of the significant features of
any individual. The color of eye may differ, the mask may
hide the face to be recognized but the underlying face
structure remains intact.
We can employ spline curves to define the basic shape of the
face and its important features like nose, chin etc. A
recognition system may take up other features along with the
basic shape of the face, apply transformations on the basic
features, compare it with the image, and recognize it in terms
of resemblances to the picture generated. Thus, we not only
formulate the feature to match but also prevent the analysis of
features in random situations. The very first step in
recognition system will be to match the equations of the
comparing faces and then on proceed to investigate at details.
Related Work
The same concept of face recognition is been accomplished by
other technologies with growing expectations on issues of
security dealing with access control, identification,
verification and authentication. First major concept is of
Eigenface [3] which works by finding eigenvectors and
eigenvalues of covariance matrix from training set images.
Similarly Fisherface which works by projecting away lighting
and facial expressions from the images while maintaining
discriminability by the optimal projection methods.[4]
Another one is the Elastic graph matching based on dynamic
link architecture. The facial features are extracted as a jet and
represented in the form of graph G with N nodes. The test
image graph is then compared to all other modal graphs. Then
is the Support Vector Machine method based on structural risk
minimization principle.[5]. Face Recognition Committee was
widelt used in neural network applications that
prophets the use of combined results of several
experts.
Proposed Technology
The concept that implements the recognition system
works with spline curves to generate the desired
equation of the face. Their application lies with the
usage of active contour models which are energy
minimizing spline curves. They are often used in
computer vision and image analysis to detect and
locate object boundaries. Also called snakes, these
contour models have solved variety of problems in
computer vision and image analysis, exemplified as
motion tracking, contours detection, intracranial
boundary detection.
Snakes basically are employed in cases where the
shape can not be defined by means of rigid
primitives. Some shapes in graphics tend to have
recognizably similar shape but do not match exactly.
Such as natural objects of same category may have
similar shape but not same. Also deformable images
have a category that changes over time such as the
facial feature like shape of lips.
How do snakes work?
Snakes may be defined as Energy minimizing spline
curves that deforms its shape to fit the local minima
which is the shape corresponding to desired image
properties. When provided with initial location
snakes may prove to be a general mechanism of
matching such deformable models to any sample
image by means of minimization of energy.
Broadly describing splines, these are flexible strips to
generate smooth curves in traditional drafting
applications. Mathematically these are approximated
curves generated out of a set of designated control
points. The spline function then approximates the
shape of the object to its best fit. Thus they can be
said to be generated as piecewise approximations of
cubic polynomial functions with zero, first and
second order continuity, just as any other Bezier
curve.
Why Splines?
There are advantages for using Spline curves:
· Control points decide the shape and nature
of the curve.
· Any number of control points can be added without
increasing the degree of polynomial.
· Splines like B- Spline lie in the convex hull of the
control points.
· Closed curves can be created with the first and
second point the same.
The parametric snake models consist basically of an elastic
curve (or surface) which can dynamically conform to object
shapes in response to internal forces (elastic forces) and
external forces (image and constraint forces). These forces can
be the result of a functional global minimization process or
based on local information. Such approach is more intuitive
than the implicit models. Its mathematical formulation makes
easier to integrate image data, an initial estimated, desired
contour properties and knowledge-based constraints, in a
single extraction process.
Mathematically, we have proved that the points generating the
curve remains almost same ( in terms of control points) . Only
but few of the points may differ reflecting the nominal
changes in the face. But due to the advantageous properties of
splines maintaining the regular shape even in case of minor
changes in terms of control points, the shape may till resemble
a individual. Thus, any disguised or altered face can be easily
recognized and changes in varying age be predicted.
Given an approximation of the boundary of an object in an
image, an active contour model can be used to find the
“actual’’ boundary. Active contour models should be able to
find the intracranial boundary in MR images of the head when
an initial guess is provided by a user or by some other method,
possibly an automated one. There have been some research
work on extraction of human face with these active contour
models [6].
The face extracted can thus be further manipulated with
desired features and requirements. This is in the form of
control points and these can be referenced for obtaining the
curve sections. These control points generated are thus the
deciding factor for generating any active contour model.
Figure-1. A standard face description in which the
shape of the lips and eye is also approximated using
splines [7].
Conclusion
In this paper we have proposed a new methodology for face
recognition. Previously this technique or method was used for
extraction of face and various features or finding the
resembling shapes. We are presenting the changes in facial
patterns over the passage of time. Our methodology is that the
basic points deciding the shape of face remains same though
the appearance may change due to health or age reasons.
Reference
[1]Daniel P Huttenlocher Ryan H Lilien, Clark F. Olson,
”View based recognition using Eigenspace Approximation to
the Housdor Measure,”
[2] Conrad sanderson , “Face processing and frontal face
verification,” IDIAP-RR 03-20, Research Report Feb 2004.
[3] M. Turk and A. Pentland, “Face recognition using
eigenfaces,” in Proc. IEEE Conf. on Computer Vision and
Pattern Recognition, 1991, pp. 586-591.
[4] P.N. Belhumeur, J.P. Hespanha, and D.J. Kriegman,
“Eigenfaces vs. fisherfaces: recognition using class specific
linear projection,” in IEEE Trans. on Pattern Analysis and
Machine Intelligence, July 1997, vol. 19, pp. 711-720.
[5] Ho-Man Tang, Michael R. Lyu and Irwin King, “Face
Recognition Committee Machine,”
[6] Yokoyama, T.; Yagi, Y.; Yachida, M.; ”Active contour
model for extracting human faces,”
[7] Marius Malciu and Fran?se Preteux, “Tracking facial
features in video sequences using a deformable model-based
approach,”
Attached Files: