Some
Past Research Projects (not a complete list)
|
Automatic Portrait Beautification
In collaboration with
Changhyung Lee (now at Samsungs), Morgan Schramm (HP), and Prof. Jan Allebach
(ECE, Purdue)
This research aims to
develop automatic methods to improve the appearance of subjects in digital
pictures. Portrait beautification is routinely performed in the fashion
industry using image editing software such as photoshop. Processing images
this way is slow and costly. Our methods allows one to process a large number
of pictures quickly and without user input.
Different
beautification steps have been considered.
·
Skin
Smoothing. Smoothing
the skin has the effect of reducing the appearance of wrinkles and blemishes
in the subject. To learn about our proposed method for automatic skin
smoothing, look at this
presentation or see the following publication:
C. Lee, M. Schramm, M. Boutin, and J. Allebach, "An Algorithm
for Automatic Skin Smoothing in Digital Portraits,” IEEE Int'l Conference
on Image Processing
(ICIP), 2009. pdf
This research was
funded by a grant from the Hewlett-Packard Company.
|
|
|
Hardware Friendly Descreening
In collaboration with
Hasib Siddiqui (now at Qualcomm) and Prof. Charles A. Bouman (ECE, Purdue).
In this project, we
developed an efficient descreening algorithm that only requires a small
number of fixed point operations. More precisely, our algorithm requires a
total of 181 additions, 18 multiplications, and 117 bitwise shifts per
trapped pixels, making it particularly attractive for hardware implementation.
For more information,
see the following publications:
·
H.S.
Siddiqui, M. Boutin, C.A. Bouman, “Hardware Friendly Descreening”, IEEE
Int'l Conference on Image Processing (ICIP), San Diego, CA, October 12-15, 2008. pdf
This research was
funded by a grant from the Hewlett-Packard Company.
|
|
|
Light-weight methods for text area
identification in natural images
In collaboration with
Prof. Ed Delp (ECE, Purdue) and Next
Wave Systems LLC.
In this project, we
developed methods for identifying the areas containing text in an image. The
methods are specially designed to be deployed on devices with a small
processor and limited energy, such as a cellular phone or a PDA. Moreover,
they perform equally well independently of the language or font used.
For more information,
see the following publication:
This research is funded
by a DARPA STTR grant.
|
|
|
Automatic translation of text in natural
scenes using a mobile device
In collaboration with
Prof. Ed Delp (ECE, Purdue) and Next
Wave Systems LLC.
In this project, we are
developing a hand-held translation device using the built-in camera of a
commercially available cellular phone or PDA. We are targeting languages that are written in a different
character set than English. Without a significant knowledge of these
languages, it is extremely difficult to obtain a translation using currently
available translation devices because of the difficulty of entering the text
by hand into the device. By taking a picture of the sign, we circumvent this
problem. See this press
release.
For more information,
see the following publication:
This research is funded
by a DARPA STTR grant.
|
|
|
Software friendly color trapping
In collaboration with
Haiyin Wang and Prof. Jan Allebach (ECE, Purdue).
For mechanical reasons,
the color planes of an image are often shiften with respect to each other
when they are printed on a color printer. This phenomena, called “color plane
mis-registration”, creates gap and halo artifacts on the printed image. Color
trapping is an image processing technique that consists in modifying the
edges of an image to hide the effects of small color plane mis-registrations.
In this project, we are developing efficient automatic color trapping
algorithms for software or firmware implementation.
For more information,
see the following publications:
This research is funded
by a grant from the Hewlett-Packard Company.
|
|
|
A novel representation method for the
shape of point configurations
In collaboration Prof. Gregor Kemper (Math,
TU Munich).
In this work, we showed
that the distribution of distances (i.e., the bag of their pairwise
distances) is a faithful representation of the shape of generic point
configurations. In other words, we showed that given any two point
configurations not belonging to an exceptional set of measure zero, there
exists a rotation, translation and reflection that maps the first
configuration onto the second if and only if the distribution of distances of
both point configurations coincide. We extended this result to point
configurations in different spaces transformed by a variety of groups. We also derived an equivalent result for the problem of
recognizing the shape of Gaussian mixtures from samples of this mixture. Our
main motivation is the problem of recognizing the shape of an object
represented by points given noisy measurements of these points.
For more information,
see the following publications:
|
|
|
Efficient Graph Representations for
browsing and indexing
In collaboration Prof. Gregor Kemper (Math,
TU Munich).
In this work, we
developed some novel representations for weighted graphs. These
representations are constructed from some simple statistics of the graph,
such as the distribution of the weights and the distribution of the sum of
adjacent weights. These representations provide us with a polynomial time
algorithm for comparing graphs. This algorithm gives the right answer in the
vast majority of time. In particular, we have shown that some of our proposed
representations yield the right answer with probability one. The motivation for this work is the
problem of browsing through a large set of graphs.
For more information,
see the following publication:
·
M.
Boutin and G. Kemper, “Lossless representation of graphs using
distributions”, submitted (2008). ps, pdf
This research is funded
by NSF grant CCF-0728929.
|
|
|
Improved conditioning of Structure from
Motion through variable elimination
In collaboration with Pierre-Louis
Bazin, Ji Zhang and Prof. Daniel G. Aliaga (CS, Purdue).
In this project, we are
concerned with the problem of reconstructing a scene from a set of images of
the scene taken from unknown viewpoints by a projective camera. We are using
symbolic equation manipulation techniques to decouple the unknown variables
(i.e., the camera pose and the scene points) in order to decrease its
complexity and improve its conditioning. In particular, we have been able to
eliminate all the camera parameters from the equations, obtaining a simple
set of degree two and three polynomial equations that provide a problem
formulation with much improved conditioning.
For more information,
see the following publications:
·
P.-L.
Bazin and M. Boutin, "Structure from Motion: a new look from the point
of view of invariant theory," SIAM Journal on Applied Mathematics, 64(4):1156-1174, 2004. ps , pdf
·
M.
Boutin, J. Zhang and D.G. Aliaga, "Improving the numerical stability of
structure from motion by algebraic elimination", IS&T/SPIE joint
symposium, Computational Imaging IV conference, San Jose, CA, January 2006. ps , pdf
·
J.
Zhang, M. Boutin and D.G. Aliaga, "Robust bundle adjustment for
structure from motion", IEEE Int'l Conference on Image Processing (ICIP), 2006. ps , pdf
·
J.
Zhang, D.G. Aliaga, M. Boutin and R. Insley, "Angle-Independent Bundle
Adjustment Refinement", 3DPVT, 2006. pdf
·
J.
Zhang, M. Boutin and D.G. Aliaga, “Variable Elimination for 3D from 2D,” IS&T/SPIE
joint symposium, Visual Communication and Image Processing conference (VCIP), San Jose, CA, Jan.-Feb.
2007. ps, pdf
This research was
funded in parts by NSF grant MSPA-MCS-0434398.
|
Last update: December 15, 2011.