3D Face Modeling and
Animation
Vuong Le
IFP group, Beckman Institute
UIUC
ECE547 Fall 2011
Contents
Motivation
3D facial geometry modeling
3D facial deformation modeling
3D facial animation
The iFace system
Speech-driven talking heads
Text-driven talking
Exercise 7 Solution
November 27, 2011
Problem 1
The answer is no. A counter example is
183
492
765
The median of the patch is 5, but it is lost no matter we rst compute the
median of each row or column.
Problem 2
Let the two new basis vectors be e1 = (1/
Exercise 5 Solution
November 11, 2011
Problem 1
Apply projection-slice theorem. The circle with radius one (D = 2) has
2DFT
D2 J1 (2)
F (, ) =
2
2
I.e.
J1 (2 u2 + v 2 )
F (u, v ) =
u2 + v 2
After stretching the circle 3 times in y -direction, the new 2DFT
Exercise 4 Solution
October 31, 2011
Problem 1
In this kind of problem, the transform we are looking for is
u
a1 a2 a0
x
v = b1 b2 b0 y
1
001
1
There are 6 unknowns, so with 3 pairs of corresponding points in xy and uv
planes the transform M can be un
Exercise 3 Solution
October 19, 2011
Problem 1
Let N be the block size. Each block takes 7+7+3+8=25 bits to encode, and
2
there are approximately 128 blocks, giving 25 1282 /N 2 bits per image. So
N2
neglecting the overhead, the compression ratio is 1282
Exercise 2 Solution
October 4, 2011
Problem 1
1
H = (0.05 log2 0.05+0.05 log2 0.05+0.1 log2 0.1+0.1 log2 0.1+0.15 log2 0.15+
0.2 log2 0.2 + 0.35 log2 0.35) = 2.5016
2
See the gure. The average bit rate per message is
0.054+0.054+0.13+0.13+0.153+0.22+0.352
Exercise 1 Solution
September 20, 2011
Problem 1
Suppose f (x, y ) F (u, v )
1.1
iii
f (x, y )
f (x, y )e2j (ux+vy) dxdy
(1)
and
f (x, y )e2j (ux+vy) dxdy
f (x, y )e2j (ux+vy) dxdy
=
(2)
f (x, y )e2j (u)x+(v)y) dxdy
=
=F (u, v )
So f (x, y ) F (u, v )
v
Motivation
Theory
Algorithms
Applications
Conclusion
Low-Rank Matrix Recovery Techniques and its
Applications in Computer Vision
Arvind Ganesh Balasubramanian
Department of Electrical & Computer Engineering
University of Illinois, Urbana-Champaign
Novembe
IMAGE WARPING
Vuong Le
Dept. Of ECE
University of Illinois
ECE547 Fall 2011
With some slides from Alexei Efros, Steve Seitz, Jilin Tu, and Hao Tang
Image Warping
image filtering: change range of image
g(x) = T(f(x)
f
g
T
x
x
image warping: change domain o
ECE547 Image Processing
Nov 3rd, 2011
Exercise 7 (ver. 1.01)
Lecturer: Prof. Thomas Huang
Due: Nov 17th, 2011
Problem 1 (Median Filter)
Is it possible to reproduce the eect of a 3 3 median lter using a 3 1 median lter on the columns
and a 1 3 median lter
1. Two-dimensional Wavelet Filter Banks
Each decomposition level of a two-dimensional wavelet transform for images consists of applying a two-channel lter bank horizontally (e.g. along each row), followed by applying that
same lter bank vertically (e.g. a
ECE547: Exercise 4
Due date: Oct. 20rd, 2011
P roblem 1
Linear warping is defined as the mapping from F(x ) to G( ) by some transformation
satisfying G(Mx ) = F(x ).
=Mx
Find the transformation matrix M of 2D warping illustrated by the following figures
ECE547 Image Processing
Sept 22, 2011
Exercise 3
Lecturer: Prof. Thomas Huang
Due: Oct 6, 2011
Problem 1 (Compression Ratio)
Compute the compression ratio that you achieved in MP3. Assume that it takes the original image
was 128 128 with 8 bits per pixel.
ECE547 Image Processing
Sep 8, 2011
Exercise 2
Lecturer: Prof. Thomas Huang
1
Due: Sep 22, 2011
Problem 2.1 (Human Code)
A set of 7 messages have probabilities: 0.05, 0.05, 0.1, 0.1, 0.15, 0.2, 0.35.
1. Calculate the Entropy per message, H.
2. Construct a
ECE547 Image Processing
Aug 25, 2011
Exercise 1
Lecturer: Prof. Thomas Huang
1
Due: Sept 8, 2011
Problem 1.1
Read lecture notes, Chapter 1 on 2D FT, and do problems 1.1 on page 1-5 (only parts iii, v, and
vii) and 1.3 on page 1-6 in the lecture notes.
2
P
ECE547: Exercise 5
Due date: Nov. 3th, 2011
Problem 5.1
=
Let
3
-1
1
60
0
-3
Projection Direction
and g ( ) its projection along a direction which makes an angle of 60 with the xaxis.
Find the 1D FT of g ( ) .
Problem 5.2
a1 x1 + a 2 x 2 + + a n x n = 1
Algorithms for Noise Removal and
MP 5 Implementation
Shiyu Chang
Prof. Thomas Huang
ECE 547
Nov 3rd , 2011
Outline
Mp5 description
Make the text in the provided image
readable
Possible way to alleviate the problem:
Non-local Mean
Bilateral Filtering
MP5 P
1
Advantages:
easier to analyze signal in pieces: divide and conquer
extracts important features
pieces can be treated in an independent manner
2
3
4
5
Multiresolution concept
Wavelet expansion are sparse compression, denosing,
Example: JPEG vs. JPEG-