DIGITAL

IMAGE
PROCESSING
(CREATIVE WORLD OF FACE MORPHING)

BY

RAMESH.S (Y5EC077) SAI KIRAN.S (Y5EC081)

[email protected] [email protected]

III/IV B.TECH III/IV B.TECH

E.C.E E.C.E

VELAGAPUDI RAMAKRISHNA
SIDDHARTHA ENGINEERING
COLLEGE
VIJAYAWADA.
ABSTRACT scope of this project, we
built up a prototypical
A study on automatic animation
face morphing is generator that can take
proposed.The an arbitrary pair of
algorithms explains the facial images and
extra feature of points generate morphing
on face and based on between them.The
these feature points, results of both inter
images are portioned and intra personal
and morphing is morphing are
performed. The subjectively
algorithms has been satisfactory.
used to generate
morphing between
images of face of
different people as well
as between images of
1.Introduction
face of individuals. To
do face morphing,
Morphing
feature points are
applications are
usually specified
everywhere. Hollywood
manually in animation
film makers use novel
industries.
morphing technologies
However,this approach
to generate special
involved computation
effects, and Disney uses
of 3N dimensional
morphing to speed up
probability density
the production of
function, N being the
cartoons. Among so
number of pixels of the
many morphing
image, and we thought
applications, we are
the approach was too
specifically interested
much computation-
in face morphing
demanding.Within the
because we believe face otherwise need to
morphing should have implement a face-
much more important finding algorithm. Due
applications than other to time-limitation, we
classes of morphing. did not study automatic
face finder.
2.Outline of
Procedures adopted 2.2FeatureFinding
Our goal was to find
4 major feature points,
namely the two eyes,
and the two end-points
of the mouth. Within
the scope of this
project, we developed
an eye-finding
algorithm that
fig 1 successfully detect eyes

2.1 Pre-Processing at 84% rate.

When getting an 2.2.1Eye-finding

image containing The fig 2 illustrates

human faces, it is our eye-finding

always better to do algorithm. We assume

some pre-processing that the eyes are more

such like removing the complicated than other

noisy backgrounds, parts of the face.

clipping to get a proper Therefore, we first

facial image, and compute the complexity

scaling the image to a map of the facial image

reasonable size. So far by sliding a fixed-size

we have been doing the frame and measuring

pre-processing by hand the complexity within

because we would the frame in a "total

 mouth as the red-most
Redness = ( R > G * 1.2 ? ) * (R>Rth?)
region below the eyes.
* { R / (G + epsilon ) }
The red-ness function is
given by
variation" sense. Then,
we multiply the Where Rth is a
complexity map by a threshold, and epsilon
weighting function that is a small number for
is set a avoiding division by
priori.Afterwards, we zero. Likewise, we can
find the three highest define the green-ness
peaks in the weighted and blue-ness
complexity map, and functions. The fig 3
then we decide which illustrates our red-ness,
two of the three peaks, green-ness, and blue-
which are our ness functions. Note
candidates of eyes, that the mouth has
really correspond to the relatively high red-ness
eyes.The similarity is and low green-ness
measured in the comparing to the
correlation-coefficient surrounding skin.
sense. Therefore, we believe
that using simple
segmentation or edge
detection techniques we
would be able to
implement an
algorithm to find the
mouth and hence its
fig 2 end points.

2.2.2Mouth-finding
After finding the
eyes, we can specify the
 points with the face
edges. Hence, totally
we have 10 feature
points for each face. In
the fig 4, the white dots
correspond to the
feature points.

2.3 Image
Partitioning

Our feature finder fig

can give us the 4
positions of the eyes
Based on these 10
and the ending points
feature points, our face-
of the mouth, so we get
morpher partitions
4 feature points. Beside
each photo into 16 non-
these facial features,
overlapping triangular
the edges of the face
or quadrangular
also need to be
regions. The partition
carefully considered in
is illustrated in the
the morphing
following two pictures.
algorithm. If the face
Ideally, if we could
edges do not match well
detect more feature
in the morphing
points automatically,
process, the morphed
we would be able to
image will look strange
partitioned the image
on the face edges. We
into finer meshes, and
generate 6 more feature
the morphing result
points around the face
would have been even
edge, which are the
better.
intersection points of
the extension line of the
first 4 facial feature
fig 4.1 fig intermediate figure
4.2 between figures 4.1 and
4.2, and the weightings
for figures 4.1 and 4.2
are alpha and (1-
alpha), respectively.
For a feature point A in
fig 4.1, and the
_ corresponding feature
point B in fig 4.2, we
are using linear
interpolation to
generate the position of
the new feature point
F:

Since the feature

points of fig 4.1 and4.2
are, generally speaking,
at different positions,
when doing morphing
between figures, the
figures have to be
warped such that their The new feature point

feature points are F is used to construct a

matched. Otherwise, point set which

the morphed figures partitions the image in

will have four eyes, two another way different

mouths, and so forth. It from images 1 and 2.

will be very strange and Images 1 and 2 are

unpleasant. warped such that their

Suppose we would feature points are

like to make an moved to the same new

feature points, and thus transformation is a
their feature points are linear mapping from
matched. In the one triangle to another.
warping process, For every pixel p within
coordinate triangle ABC, assume
transformations are the position of p is a
performed for each of linear combination of
the 16 regions A, B, and C vectors.
respectively. The transformation is
given by the following
2.4CoordinateTrans equations,
formations

There exist many

coordinate
transformations for the
Here, there are two
mapping between two
unknowns, Lambda1
triangles or between
and Lambda2, and two
two quadrangles. We
equations for each of
used affine and bilinear
the two dimensions.
transformations for the
Consequently,
triangles and
Lambda1 and
quadrangles,
Lambda2 can be
respectively. Besides,
solved, and they are
bilinear interpolation is
used to obtain q. i.e.,
performed n pixel
the affine
sense.
transformation is a one-
to-one mapping
2.4.1 Affine
between two triangles.
Transformation

Suppose we have two

triangles ABC and
DEF. An affine
2.4.2 Bilinear can be solved, and they

Transformation are used to obtain

q.Again, the Bilinear
transformation is a one-
to-one mapping for two
quadrangles.

Suppose we have two

2.5 Cross-
quadrangles ABCD and
EFGH. The Bilinear
Dissolving
transformation is a
After performing
mapping from one
coordinate
quadrangle to another.
transformations for
For every pixel p within
each of the two facial
quadrangle A Bilinear
figures, the feature
transformation is given
points of these figures
by the following
are matched. i.e., the
equations, BCD,
left eye in one figure
assume that the
will be at the same
position of p is a linear
position as the left eye
combination of vectors
in the other figure. To
A, B, C, and D .
complete face
Bilinear equations are
morphing, we need to
do cross-dissolving as
the coordinate
transforms are taking
place. Cross-dissolving
is described by the
following equation,
There are two
unknowns u and v.
Because this is a 2D
problem, we have 2
equations. So, u and v
where A,B are the pair 2. Perform coordinate
of images, and C is the transformations on the
morphing result. This partitioned images to
operation is performed match the feature
pixel by pixel, and each points of two images.
of the color components Here, we are matching
RGB are dealt with the eyes and the mouths
individually. for two images. We can
find that Ally's face
The following example
becomes longer than
demonstrates a typical
lions face .
morphing process.

1. The original images

of Ally and Lion, scaled
to the same size. Please
note that the distance
___
between the eyes and
the mouth is
significantly longer in
the lion's picture than
in Ally's picture.

3. Cross-dissolve the
___ two images to generate
a new image. The new
face possesses the
features from both
Ally's and the lion's
faces.
_________ __

3.Recent Evolutions

(1) Morphing
between faces of
different cases: (2) Morphing
between different
- human and animal
images of the same
(lion)
person :
- man and man
- man and woman We want to interpolate
the intermediate
In the following, the
expressions or poses by
very left and very right
morphing.
images of each row are
original images, and the
Happy <===
intermediate ones are
===> Angry
synthesized morphed
images.

4. Conclusion
 An automatic face feature points we can
morphing algorithm is specify on the faces, the
proposed. The better morphing results
algorithm consists of a we can obtain. If we
feature finder followed can specify all the
by a face-morpher that important facial
utilizes affine and features such as the
bilinear coordinate eyes, the eyebrows, the
transforms. We believe nose, the edge points of
that feature extraction the mouth, the ears,
is the key technique and some specific
toward building points of the hair, we
entirely automatic face are confident that we
morphing algorithms. can generate very
Moreover, we believe smooth and realistically
that the eyes are the looking morphing from
most important one image to another.
features of human
faces.We hence 5. References
achieved an 84% of eye
[1] Martin Bichsel,
detection rate. Also, we
"Automatic
proposed red-ness,
Interpolation and
green-ness and blue-
Recognition of Face
ness function and
Images by
illustrated how we
Morphing",
would be able to find
proceedings of the 2nd
the mouth based on
international
these functions. Also,
conference on
we demonstrated
automatic face and
that face morphing
gesture recognition,
algorithm can help
pp128-135
generate
[2] Jonas Gomes et al.
animation.Ideally
"Warping and
speaking, the more
morphing of graphical
objects ", Morgan
Kaufmann Publishers
(1999).